افزايش روزافزون كاربرد سيست مهاي مخابراتي و راداري په نباند و حساسيت اين تجهيزات به تداخ لهاي الكترومغناطيسي (EMI) ، نياز مبرمي به توسع هي جاذبهاي موج پيشرفته با قابليت جذب بالا و پهناي باند عملياتي گسترده ايجاد كرده است. جاذبهاي فراماد هاي (Metamaterial Absorbers - MMA) به دليل امكان كنترل بيسابق هي پارامترهاي مؤثر الكتريكي و مغناطيسي از طريق طراحي هندسه، پاسخي ايد هآل براي اين نياز فناورانه هستند. با اين حال، اغلب اين ساختارها ذاتاً باند باريكي دارند . هدف اصلي اين پاياننامه، طراحي، بهين هسازي و تحليل يك جاذب فراماد هاي صفح هاي فوقنازك و په نبان دو قابل تنظيم براي كاربرد در محدود هي فركانسي X/Ku ( ??.?? تا ??.? گيگاهرتز( است. در اين راستا، ابتدا يك ساختار پايه مرور شد و سپس از طريق يكرويه گا مب هگام سيستماتيك، با اعمال تغييرات هدفمند در هندس هي سلول واحد، نظير افزودن حلق ههاي ه ممركز، ايجاد و تنظيم شكا فهاي تشديدي و معرفي مسيرهاي مارپيچ، ساختار اصلي توسعه يافت . مكانيسم عملكرد مبتني بر ايجاد همزمان رزونان سهاي الكتريكي و مغناطيسي و كوپلينگ مؤثر بين آ نها است كه منجر به تطبيق امپدانس بهينه با فضاي آزاد و در نتيجه كاهش چشمگير بازتاب ميشود . براي تحليل عملكرد، از شبي هسازي الكترومغناطيسي تما مموج در نرمافزار CST Studio Suite استفاده شد و پارامترهاي پراكندگي، توزيع ميدان و جريان سطحي به دقت بررسي گرديد. در مرحلهي بعد، اثر آراي هسازي و كوپلينگ متقابل بين سلولها با بررسي دو آرايه با فواصل سلولي متفاوت مورد مطالعه قرار گرفت . دستاوردهاي كليدي اين پژوه شعبارتند از : ? ( دستيابي به يكسلول واحد بهينه با ضخامت تنها ?.??? ميل يمتر معادل تقريبي ? ? در فركانس مركزي كه جذب بالاي ??? را در پهناي باند ?.?? گيگاهرتزي ارائه م يدهد و در نقاط اوج به بازدهي حدود ??? ميرسد. ? ( نشان دادن اين كه كوپلينگ متقابل در آراي ههاي فشرد ه ميتواند به عنوان يك پارامتر طراحي مثبت عمل كرده و منحني پاسخ را هموارتر نمايد . ساختار پيشنهادي به دليل سازگاري با فناوري ساخت PCB و عملكرد پايدار در زواياي تابش مختلف، گزينهاي عملي و اميدبخش برا ي كاربرد در سيستمهاي راداري، مخابراتي نوين و فناور يهاي كاهش سطح مقطع رادار ي (RCS) محسوب ميشود  

امروزه شبكه‌هاي جهاني وب تبديل به يكي از ابزارهاي مورد نياز بشر شده‌اند كه توسط كاربران بسياري در سراسر جهان مورد استفاده قرار مي‌گيرند. مساله‌اي كه در اين حوزه وجود دارد گستردگي بسيار زياد اينترنت و مطالب آن است كه اين گستردگي روز به روز و با سرعت بسيار زياد در حال افزايش است. اكنون يكي از مشكلاتي كه پديد مي‌آيد، اتلاف وقت كاربران براي دستيابي به كالا ها و خدمات مورد نياز آن‌ها است و ممكن است در بسياري مواقع كالا يا خدمت مورد نياز خود را پيدا نكنند، در نتيجه ارائه و پيشنهاد كالا يا خدمات مناسب به كاربران در زمينه هاي مختلف مطابق با نيازها آن‌ها امري بسيار حياتي محسوب مي گردد و يكي از روش هاي بسيار پركاربرد براي اين مساله استفاده از سيستم هاي توصيه‌گر مي‌باشد. ?سيستم‌هاي توصيه‌گر به منظورجلوگيري از اتلاف وقت كاربران، محصولاتي را به آنها پيشنهاد مي‌كنند كه به احتمال زياد مورد علاقه آن‌ها هستند و هنوز آن‌ها را نديده‌اند. الگوريتم پالايش گروهي به عنوان يكي از معروف ترين و پركاربرد ترين الگوريتم‌ها براي پياده سازي يك سيستم توصيه‌گر شناخته مي‌شود. سيستم‌هاي مبتني بر اين الگوريتم بر اساس سوابق جستجو و ابراز علاقه‌مندي كاربر به كالا‌ها و خدمات مختلف و با توجه اطلاعات دريافتي از ديگر كاربران با علاقه‌مندي‌هاي مشابه به كاربر هدف پيشنهادات جديدي را ارائه مي‌كنند. سيستم‌هاي توصيه‌گر اغلب با چالش مهمي به نام تنكي داده‌ها مواجه هستند، زيرا ماتريس‌هاي تعامل كاربر–اقلام در پياده‌سازي‌هاي واقعي معمولاً بيش از ??? تنك هستند. اين مسئله تأثير منفي بر دقت و كارايي توصيه‌ها دارد، به‌ويژه براي كاربران جديد و محصولات خاص يا كم‌تعامل. در حالي كه روش‌هاي موجود تلاش مي‌كنند با افزودن اطلاعات جانبي يا تغيير در طراحي سيستم اثر تنكي داده را كاهش دهند، اغلب مشكل اصلي يعني كمبود داده‌هاي تعاملي را ناديده مي‌گيرند. در اين پژوهش، يك چارچوب جديد معرفي مي‌شود كه از يك مدل ديفيوژني براي توليد امتيازدهي‌هاي مصنوعي با كيفيت بالا در سيستم‌هاي توصيه‌گر استفاده مي‌كند. به طور مشخص، از يك مدل ديفيوژني طراحي‌شده براي داده‌هاي جدولي استفاده مي‌كنيم تا توزيع مشترك و پيچيده‌ي سه‌تايي‌هاي كاربر–اقلام–امتياز را ياد بگيرد و سپس امتيازهاي مصنوعي توليد كند كه از نظر آماري سازگار هستند. براي تضمين كيفيت امتيازهاي توليدشده، يك روش نوآورانه براي شناسايي نويز بر اساس تحليل الگوهاي رفتاري پيشنهاد مي‌كنيم. اين روش امتيازهايي را كه با ترجيحات كاربران و ويژگي‌هاي آيتم‌ها همخواني ندارند، شناسايي كرده و حذف مي‌كند. براي ارزيابي كارايي چارچوب پيشنهادي، از مدل‌هاي پالايش گروهي سنتي و پالايش گروهي مبتني بر شبكه‌هاي عصبي استفاده شده است. آزمايش‌ها روي دو مجموعه‌داده واقعي با سطوح مختلف تنكي داده (با نگهداشت داده از 5% تا ???%) انجام شد و بهبودهاي قابل‌توجهي را نشان داد. به طور خاص، پالايش گروهي سنتي مي‌تواند خطاي RMSE را تا ??% كاهش دهد و نيز پوشش امتيازدهي را تا ?? % افزايش دهد. همچنين، پالايش گروهي عصبي پاسخ‌هاي دقيق‌تر و ظريف‌تري ارائه مي‌دهد و زماني بيشترين كارايي را دارد كه نسبت افزايش داده پايين باشد. اين موضوع نشان مي‌دهد كه مدل‌هاي مختلف به انواع متفاوتي از افزايش داده نياز دارند.   

   The rapid growth of social commerce as an emerging branch of e-commerce has created extensive opportunities for businesses and users. However, this expansion has been accompanied by a significant rise in cybercrime incidents, exposing users to serious security challenges. The increasing sophistication of malicious behaviors, the exploitation of emerging technologies by cyber attackers, users’ limited cybersecurity awareness, and the absence of adequate security infrastructures are among the key factors that heighten the vulnerability of social commerce environments. Accordingly, the primary objective of this study is to identify the major challenges contributing to cybercrime occurrence in social commerce and to propose effective countermeasures with a specific emphasis on user empowerment. This research adopts a mixed-method approach (qualitative–quantitative). In the qualitative phase, a systematic content analysis of authoritative sources along with expert consultation through the Delphi method was conducted to identify the most influential challenges and components. Subsequently, in the quantitative phase, the extracted factors were evaluated using a structured questionnaire and statistical analysis to determine the significance and impact of each variable on social commerce security. The findings reveal seven key components—user awareness and training, security infrastructures and technologies, user trust and interaction, legal and policy-related challenges, technical challenges and specific cyberattacks, user empowerment, human-resource-related factors, and emerging technologies—that play a critical role in both enabling and mitigating cyber threats. The results indicate that user empowerment across three core dimensions—enhancing digital and security literacy, improving threat-detection capabilities, and promoting safe behavioral practices in online social interactions—can substantially reduce the likelihood of users becoming victims of cybercrime. Furthermore, strategies such as developing targeted educational programs, designing user-centric security mechanisms, strengthening protective regulations, and employing advanced technologies such as artificial intelligence for threat monitoring are essential for establishing a secure social commerce ecosystem. Ultimately, the conceptual model proposed in this research can serve as a practical framework for platforms, policymakers, and businesses operating in the field of social commerce and contribute meaningfully to reducing cybercrime and improving user security.

The primary aim of thisresearch is to design and implement an intelligent and efficient framework forintrusion detection in Internet of Things (IoT) devices using a novel FederatedLearning (FL) approach. With the rapid growth of IoT applications in domainssuch as healthcare, industry, agriculture, and smart cities, a massive amountof data is generated by connected devices. Ensuring the security and privacy ofthis data has become a critical challenge. Traditional centralized intrusiondetection systems (IDSs) are no longer suitable due to their high communicationoverhead, limited device resources, and hardware heterogeneity. To address these challenges,this thesis introduces a new framework called ASA (Adaptive Smart Agent). ASAemploys an adaptive agent layer that monitors device resources and dynamicallyclusters IoT devices based on their computational power, memory capacity, andbandwidth. For each cluster, an appropriately scaled learning model isassigned. The training process is performed locally on devices, and only modelupdates are transmitted to the central server, thereby reducing communicationcosts and preserving user privacy.Experimentalevaluations on benchmark IoT datasets demonstrate that ASA significantlyoutperforms conventional FL-based methods in terms of detection accuracy,communication efficiency, and participation fairness. It effectively mitigatescritical issues such as device dropouts, non-IID data distribution, and networkinstability, while maintaining robustness and stability in heterogeneousenvironments.Theresults highlight that the proposed ASA framework enhances the accuracy andscalability of IoT intrusion detection systems while ensuringprivacy-preserving distributed learning. Future work can focus on acceleratingmodel convergence, improving fault tolerance, and integrating ASA with edge andfog computing infrastructures to enable real-world deployment in large-scaleIoT ecosystems.  

   Approximate computing is a promising approach for high-performance, and low-energy computation in inherently error-tolerant applications. This study proposes an approximate adder comprising a constant-truncation block in the least significant part and several non-overlapping summation blocks in the more significant parts of the adder. The carry-in of each block is supplied using the most significant bit of one of the input operands from the earlier block. In the most significant block, two more-precise approaches are used to generate candidate values for the carry-in. The final value of the carry-in for this block is selected based on the values of the input operands. In fact, the proposed approximate adder is input-aware, and dynamically adjusts its operation in one or two cycles to improve accuracy while limiting the average delay. The experimental results indicate that the proposed adder has a better quality-effort tradeoff than state-of-the-art approximate adders. Different configurations of the proposed adder improve delay, energy, and the energy-delay product (EDP) by 78%, 72% and 87% respectively, when compared to state-of-the-art approximate adders, all without any loss in accuracy. Additionally, the efficiency of the proposed adder is confirmed in both image dithering and stock price prediction through regression.

   With the rapid development of smart cities and the growing need for accurate and real-time analysis of road infrastructure, the design of AI-based systems capable of perceiving, analyzing, and recording environmental data has become increasingly crucial. In this regard, the present study focuses on the design and implementation of an innovative system for the automatic detection, tracking, and localization of traffic signs. This system not only pushes technical boundaries but also makes a significant contribution to the localization of traffic-related data. The advanced YOLOv9 model is employed for precise traffic sign detection, while the powerful ByteTrack algorithm ensures continuous tracking. What truly distinguishes this research is the novel application of federated learning using the FedAvg algorithm—implemented for the first time in the domain of traffic sign recognition. This method enables the training of models on heterogeneous datasets, including two distinct subsets, DFG and Mapillary, without requiring physical data aggregation. This approach not only preserves data privacy but also significantly enhances the generalization capability of the model. On the data side, the study introduces a rich and unprecedented dataset comprising 14,111 images and over 19,000 traffic sign instances across 118 classes. The data was collected over two years in varying temporal conditions (morning, noon, evening, night) and all four seasons, spanning urban, rural, and interurban areas across the country using mobile phone cameras. The images were meticulously annotated using the Makesense tool in both YOLO (.txt) and Pascal VOC (.xml) formats. The system’s performance, evaluated through 6-Fold Cross Validation, demonstrates its high accuracy, achieving a remarkable mAP50 of 95.66%. This not only reflects the model's robustness in real-world conditions but also shows a clear advantage over various versions of YOLO from v5 to v11. The initial idea for this project stemmed from a proposal by Tehran Municipality to design a digital map of traffic signs. However, the outcomes of this research go far beyond a municipal application and offer valuable tools for navigation systems, intelligent vehicles, spatial analytics, and the development of a national traffic sign map in Iran. This work presents a seamless integration of cutting-edge technology, locally-driven data, and modern AI architectures to pave the way for a smarter and safer future on the country’s roads.

includes 2605 training samples and 1321 test samples. The labeling of these data was done with two classes positive and negative bythree annotators using the majority voting method. In this study, five different architectures, namely Bert-fa-zwnj-base, Bert-fa-base-uncased, LSTM, GRU, and Distil-bert, were employed for sentiment analysis, and these models were evaluated with two optimizers, SGD and Adam. The results indicate that the Bert-fa-base-uncased model performed the best on both datasets, achieving an accuracy of 93% on the Twitter dataset and 80% on the Instagram dataset. Furthermore, the Adam optimizer outperformed SGD. This research demonstrates that the use of deep learning-based models, especially Bert-fa-base-uncased, can effectively perform sentiment analysis on Persian texts with high accuracy and efficiency, processing data generated on widely used platforms such as Instagram and Twitter effectively.  

رايانش مه يك زيرساخت توزيع شده با امكان ارتباط، ذخيره‌سازي و محاسبه در لبه يك شبكه محلي و بسيار پوياست. فاصله زياد سرويس­گيرنده­هاي يك محيط محلي با ابر و همچنين تعداد بسيار بالاي درخواست‌ها از ساير محيط­ها كه حساس به تأخير هستند مشكلاتي را در ارائه خدمات ابري به وجود آورده است. درنتيجه استفاده از قابليت محاسباتي منابع بيكار محلي و نزديك به دستگاه‌هاي انتهايي همانند خودروهاي با/بدون سرنشين و ايجاد يك شبكه ad-hoc تحت عنوان رايانش مه وسايل نقليه سبب كاهش ارسال درخواست‌ها به ابر و همچنين كاهش زمان پاسخ مي‌شوند. با اين ‌وجود محدوديت منابع در رايانش مه وسايل نقليه در مقايسه با ابر باعث ايجاد مشكلاتي از قبيل يافتن منابع آزاد از نظر توان محاسباتي و همچنين دسترس‌‌پذيري منابع در ارائه سرويس مطلوب به مشتري­ها مي­شود. درنتيجه تلاش براي پيش­بيني درست ميزان منابع درخواستي هر وظيفه مي­تواند از هدر رفتن منابع محدود گره­هاي مه جلوگيري كند كه اين امر نيازمند روش‌هايي از قبيل يادگيري ماشين است تا بر اساس درخواست/پاسخ‌هاي دستگاه‌هاي انتهايي بتواند رفتار محيط را تا حدودي ياد گرفته و جهت رسيدن به كيفيت مطلوب سرويس­دهي، مقادير مناسبي از منابع را در اختيار آن­ها قرار دهد. در اين پژوهش با استفاده از يادگيري تقويتي عميق QL روشي براي برنامه‌ريزي و پيش‌بيني منابع مورد‌نياز يك مشتري خودرو هوشمند با معماري سه‌لايه رايانش خودرويي در بهينه‌سازي تخصيص منابع و بهبود عملكرد كلي سيستم ارائه شده است.اين روش با استفاده از قابليت‌هاي هوش مصنوعي و يادگيري تقويتي، رويكردي پويا و تطبيقي براي مديريت منابع در يك محيط محاسباتي مه ارائه مي‌دهد. دو الگوريتم اصلي براي مسئله پيش‌بيني و تخصيص منابع در اين تحقيق پيشنهاد شده است. در انتها بر مبناي روش پيشنهادي از ابزارها و مجموعه داده­هاي مناسب جهت ارزيابي استفاده مي‌شود. داده‌هاي مورداستفاده هم مي‌توانند يك بازه‌اي مشاهده شده از دنياي واقعي باشند و هم مي‌توانند از طريق ابزارهاي شبيه‌سازي مانند Matlab توليد شوند. ديتاست مورداستفاده شامل وضعيت‌هاي خودروهاي كلاينت، درخواست‌هاي آنها، گره‌هاي مه، تحركات آنها و وظايف درخواست شده از سمت كلاينت‌ها در يك بازه زماني خاص است. يافته‌هاي كليدي اين مطالعه نشان مي‌دهد كه يادگيري تقويتي QL مي‌تواند به طور مؤثري ميزان متناسب تخصيص منابع را با يادگيري از تجربيات گذشته و تصميم‌گيري آگاهانه پيش‌بيني كند. با آموزش و به روزرساني مستمر عامل يادگيري Q، سيستم مي‌تواند با شرايط متغير سازگار شود و تصميمات تخصيص منابع را بر اساس اطلاعات بلادرنگ اتخاذ كند. همچنين نتايج آزمايش‌ها اثربخشي روش پيشنهادي را در بهينه‌سازي تخصيص منابع نشان مي‌دهد. عامل يادگيري تقويتي QL اقدامات بهينه‌اي را ياد مي‌گيرد كه مصرف منابع را به حداقل مي‌رساند درحالي‌كه الزامات عملكرد سيستم مه را برآورده مي‌كند. اين منجر به بهبود كارايي، كاهش تأخير و افزايش قابليت اطمينان سيستم مي‌شود.   

 Detecting brain tumors from the levels of MRI images is one of the biggest challenges in the world of artificial intelligence and engineering sciences Medicine goes to the number. Brain tumors, which can lead to the death of people with their growth, need to be staged Identify, identify. There are two main categories of tumors, which include benign and malignant tumors. made An intelligent medical diagnosis system in the field of brain tumor diagnosis from the levels of MRI images is one of the important parts In medical engineering science, it is numbered, which can help doctors in the early detection and identification of tumors and then diagnosis. Care and maintenance of people until full recovery is helpful. Glioma is the most common malignant brain tumor with grades It is different, which greatly determines the survival rate of patients.Tumor segmentation and grading using contrast-enhanced imaging Magnetic resonance imaging (MRI) is essential for diagnosis and treatment planning. To achieve this clinical need, a feature Part of Convolutional Neural Networks (CNN) based on UNet network for tumor segmentation and transmission Then, based on the pre-trained convolution of the Vgg16 base and a perfection classifier for tumor grading, it is developed. Segmentation and gradation models from the same pipeline T1 - Coping, weak liquid damping inversion recovery (FLAIR) and post-contrast T1 MRI images of 110 low-grade glioma patients (LGG) for teaching and evaluating the use of do Dice similarity coefficient (DSC) and tumor detection accuracy obtained by the segmentation model are 0.84 and 0.92, respectively.Model Grading LGG with accuracy, sensitivity and specificity of 89.0, 87.0 and 92.0 respectively at the level of MRI images and 95.0, 97.0 and 98.0 classifies patients as benign and malignant. This work has the potential to use deep learning MRI images to provide a non-invasive tool for simultaneous and automated tumor segmentation, diagnosis and tumor grading shows for clinical applications.

Sleep is a state of reduced mental and physical activity and is vital for part of daily activities. When sleep is insufficient, several problems such as learning disorders and increased risk of stress diseases such as mood disorders and cardiovascular diseases appear. The older population, adults over 60, have more problems with sleep and sleep disorders. They are more prone to sleep disorders, such as insomnia. These problems together can lead to an increased risk of other diseases. A common condition among the elderly is dementia. Lack of sleep and bad sleeping habits are risk factors for dementia, however, sleep is not the only risk factor for dementia. Education, age and gender of people are other factors that play a role in the risk level of a given person. Since data is available on the older population, we can make inferences about different aspects of their lives. Machine learning is a program that uses experience to learn and make predictions based on its experiences. These programs can find the patterns in the data or explain the relationship between them. We use ML to study the relationship between sleep and dementia. We also use other features such as methodological approaches and other aspects of sleep such as REM sleep, so using machine learning with different features is useful in other aspects of sleep. The current thesis evaluates different methods to investigate sleep disorders on dementia using five machine learning algorithms (gradient boosting, logistic regression, Gaussian smoothing, random forest and support vector machine). Data on the older population (60+) in Sweden from the Swedish National Study on Aging and Care - Blacking = 4175 (number of samples) Algorithm from 10-fold stratified cross-validation to obtain results, including Brier score to check accuracy And feature importance is used to investigate factors affecting dementia. Algorithms use 16 features that are based on personal factors and sleep disorders.  

Abstract Objective: Due to the increasing growth of image transmission in computer networks, it is very important to provide a suitable level of security to protect these images, which can be ensured by using different encryption methods. Image encryption methods based on chaos theory are known as a more effective and safer solution in image encryption due to the unique characteristics of chaos functions, such as sensitivity to initial values and parameters and high scrambling power. This study was conducted with the aim of determining the optimal chaotic mapping for encryption of four different groups of images including face images, fingerprint images, satellite images and medical images and increasing the encryption speed. Research method: First, texts and articles related to image encryption were studied using chaotic mappings. Using these studies, 11 one-dimensional and two-dimensional chaotic maps were investigated. In the implementation phase, 40 images were encrypted with these 11 maps using the Python programming language. Then, in the evaluation stage of encrypted images, the encryption quality was checked with the help of criteria such as image histogram and correlation between image pixels. After the evaluation stage, it was determined that for the encryption of each of these image grou   which mapping is more appropriate? In the end, the encryption speed was increased by using parallelization techniques. Findings: The result of this study was to determine the appropriate chaotic mapping for encryption of each of the four image groups and the parallelization of chaotic key generation. Also, the chaotic function of sinusoidal mapping was improved by making changes in the equation of this mapping. After analyzing the encrypted images, it was determined that Logistic 2, Logistic 3, Duffing, and Sinusoidal mappings are the optimal mappings for face, medical, fingerprint, and satellite image encryption, respectively. It was also found that chaotic quadratic mapping has the highest speed of generating chaotic keys. Conclusion: One of the available methods to ensure the security of images during transmission in computer networks is image encryption. In image encryption, one of the important steps is to generate encryption keys. Pseudo-random keys can be generated by using chaotic functions. There are different types of chaotic functions that it is better to choose a suitable function for image encryption according to the type of image. The use of chaos functions increases the security factor of image encryption; but it usually requires a lot of calculations. This volume of calculations can reduce the encryption speed. To solve this problem, different parallelization methods can be used. Keywords: image encryption, chaos functions, optimal mapping, parallelization   

Verifying a person's identity using handwritten signatures is challenging in the presence of a skilled forger, where the forger has access to the person's signature and deliberately tries to imitate it. In offline (static) signature verification, the dynamic information of the linear signature process is lost, and it is difficult to design good feature extractors that can distinguish between genuine signatures and skilled forgeries. A signature is a handwriting of people that has special features and makes each person's signature unique, so a system can be designed to recognize people's signatures and authenticate their identity by means of signatures. One of the machine learning methods that has the appropriate accuracy to detect such projects is convolutional neural networks. In this research, we combined the deep convolutional network model with the federated learning approach, which provides proper accuracy in signature detection. This model recognizes professional forgery signatures with an accuracy of more than 91% and random forgeries with an accuracy of about 97%.  

Nowadays, social networks have become an inseparable part of people's lives; As far as most people use these networks to express their feelings about all aspects of life. Because understanding and analyzing these emotions has many applications, such as customer relationship management, checking and observing people's mental health, identifying public emotions caused by a national, global or political event, identifying criminals and improving the efficiency of responsive robots, etc.; Identifying the intensity of emotion from the texts of social network users is considered a very practical and important issue. In this project, we are trying to find the most accurate model for predicting the intensity of emotion from social network tweets with the help of different natural language processing methods, using regression-based machine learning models and neural networks. In this research, we first used basic methods such as Bag of Words, Word2Vec, GloVe and TF-IDF and obtained accuracy on the data of the SemEval2018 Task1 EI-Reg competition, then using modern methods such as GPT2 and model Based on BERT, we perform various tests on these data in order to reach the best possible result in terms of Pearson correlation. The result obtained from the methods used, which is a combination method of different models, is equal to 0.82, which compared to the previous works in this research and the teams participating in SemEval competition is the best result.

Accurate and timely diagnosis of brain tumors is essential for effective treatment of this disease. The choice of treatment method depends on the level of the tumor at the time of diagnosis, the type of pathology and its grade. Glioma brain tumor is the most common type of primary brain tumor and all of them originate from glial cells that surround neurons. In diagnosing this type of disease, computer-aided diagnosis methods have helped neurologists in various ways. Recent works in this field have led to improved efficiency with the emergence of the concept of deep learning. Computer-aided recognition systems approaches include preprocessing, segmentation, feature analysis (feature extraction, feature selection, and feature verification) and >   In this thesis, methods based on image processing, machine learning, and deep learning have been used to identify glioma brain tumors and >The dataset used in this research is Brats2018, which includes 210 MRI images of high-grade glioma tumors and 75 images of low-grade glioma tumors. In the first proposed method, image segmentation was done using the combination of K-Means clustering algorithm and Coyote optimization algorithm, as well as different feature extraction methods, including texture feature extraction using local binary patterns method and deep feature extraction. From the trained neural network and the pre-trained VGG16 network, among the mentioned methods, the extraction of deep features resulting from the precise adjustment of the pre-trained VGG16 network brought the best results. This proposed method reached an acceptable accuracy of 99%. In the second proposed method, the clustering centers were optimized using the coronavirus algorithm, and the previous feature extraction methods were also implemented for this method, and the best performance related to feature extraction was through a trained neural network. be In this process, we reached 99.80% accuracy. Keywords: Glioma Brain Tumors, Clustering, Coyote Meta-heuristic Algorithm, Coronavirus meta-heuristic algorithm   

   Due to the growth of the use of electronic devices such as electronic amplifiers, wireless devices, etc., power dividers have gained special importance, and the most important of these dividers are Wilkinson and Gysel, which are widely used. take The main task of dividers in the electronic circuit is power division, which is evaluated in symmetrical and asymmetrical divider models today.    The problem discussed in this thesis is to pay attention to the reduction of the physical size that affects the size of the electric circuit, that the reduction of the designed dividers reduces the dimensions of the circuit. It increases the quality of the output wave and also increases the conduction bandwidth, which plays a significant role in the selection of dividers in devices, which is investigated using microstrip technology today. In this thesis, ADS software has been used to simulate a Gysel divider, which has been obtained by reducing the dimensions, removing additional harmonics, and providing proper isolation, and good results have been obtained.

   Abstract: Today, with the increasing expansion of social networks, users have access to the opinions and views of other people. These opinions often contain valuable information that can be analyzed to understand people's tastes and tendencies and to identify their positive, negative or even neutral opinions on various issues. But since the volume of these data and the speed of their production is surprisingly high, analyzing it by humans is a difficult, time-consuming and practically impossible task; Therefore, there is a need for a system that can automatically analyze comments. Sentiment analysis is the solution to this problem. Sentiment analysis is a process that is able to discover people's views, attitudes and feelings from their writings. Sentiment analysis or opinion analysis is a subset of text mining and natural language processing, the purpose of which is to automatically extract users' views on various issues. Microblog is a type of social network where users try to share their short texts with others. Twitter is one of the most popular microblogs in which the maximum size of each tweet is 280 characters, and this feature has made Twitter a suitable platform for knowing the opinions of users. In this thesis, sentiment analysis has been done on 7306 Farsi tweets extracted from the Twitter social network on the topic of Corona vaccine. For this purpose, tweets were considered in three >Keywords: text mining, sentiment analysis, corona vaccine, natural language processing, machine learning, deep learning, Twitter social network

Abstract The term cryptocurrency is an emerging topic in today's world, which has created a revolution in our vision in the field of investment and has caused changes in the world's financial systems. Cryptocurrency is a digital currency that uses blockchain technology with secure encryption. Every change can have advantages and disadvantages, cryptocurrencies are no exception to this rule, and along with their advantages, they can also have disadvantages for the economy of any society, so that due to the decentralization of these currencies, traditional monetary systems and the capital market of each They can influence a society. Therefore, due to the importance of the issue, the need to understand public opinion and analyze people's opinions in this regard increases. To understand the opinions and views of people about different topics, you can take help from social networks because they are a rich source of opinions. The Twitter social network is one of the main platforms where users discuss various topics, therefore, in the shortest time and with the lowest cost, the opinion of the community can be measured on this social network. Twitter Sentiment Analysis (TSA) is a field that analyzes the sentiment expressed in tweets. Considering that most of TSA's research efforts on cryptocurrencies are focused on English language, the purpose of this research is to investigate the opinions of Iranian users on the Twitter social network about cryptocurrencies and provide the best model for >  

Network Functions Virtualization of architecture means providing various network services without the need for hardware and not depending on it. Network Functions Virtualization is a new field in the network, with the help of which hardware devices can be implemented in virtual and software form. Network Functions Virtualization improves network functions such as: proxies, firewalls, load balancing, etc. In other words, using virtualization technology, this architecture is able to convert hardware devices into software modules known as virtual network functions and provide the desired service to the user. Providing the service requested by the user in the network is done by a sequence of virtual network functions, which are known as service functions chain. One of the main challenges in the development of network functions virtualization architecture is the allocation of resources to the requested network services in network infrastructures based on network functions virtualization, this challenge is called network function virtualization resource allocation problem. Therefore, in this research, the problem of allocating resources to virtual network functions in Network Functions Virtualization architecture has been solved by using mathematical programming techniques. In this research, a multi-objective mixed integer linear programming model is presented for the problem of resource allocation to virtual network functions. In this model, constraints related to the resource capacity of nodes and connections and delay constraints are desired. Also, the objective functions in this research are: maximum flows accepted in the network, reduction of resource costs of nodes (including: the number of CPU cores and the amount of memory), reduction capital costs, reduction operational costs and checking execution time. These constraints and objective functions are expressed precisely and explicitly by mathematical functions. The proposed mathematical model is implemented and solved with the Cplex solver. To evaluate the proposed mathematical model, several different topology are considered. The optimal cost is evaluated under changing parameters such as the length of service functions chain, the number of flows, the length of flows, the amount of resources of nodes, the number of nodes and the number of virtual network functions. And finally, the increase in execution time is checked by changing the number of nodes and the number of virtual network functions. The numerical results of this research show the effectiveness of the model in resources allocation to virtual network functions. Keywords: Network Functions Virtualization architecture, Virtualized Network Functions, Resource allocation, Mathematical programming, Mixed integer linear programming   

   Normal 0 false false false EN-US X-NONE FA

   Today, with increasing in the demand for electric energy consumption in the world and the limitation of the use of fossil fuels, the increasing attention of energy producers to the use of clean and renewable sources such as solar energy and wind energy has been attracted, and producers have introduced these resources as alternatives with higher reliability and quality, and better environmental and economic considerations. The use of renewable energy sources to generate electricity requires distributed generation systems in the form of microgrids. Microgrids are composed of distributed generation sources, local loads and energy storage system, and they are able to supply the required load of the system in the mode of direct connection to the national grid or island connection. AC microgrids, DC microgrids and AC/DC hybrid microgrids, are three main structures of microgrids. DC microgrids have received special attention due to their simpler control compared to AC microgrids and AC/DC hybrid microgrids. However, the lack of inertia in the microgrids, especially the DC microgrid in island mode, has caused them to have less resistance and stability margin against disturbances and fluctuations. One of the main issues in DC microgrids is voltage swing control of these microgrids. In this thesis, the design and simulation of an island mode DC microgrid which includes solar energy source, energy storage system and constant power loads is discussed. In this system, a virtual DC machine is used to improve damping and generate inertia. In order to have a stable and optimal performance of DC microgrid against oscillations and also to improve and increase the inertia and damping of the microgrid, a control method based on robust control H?   has been presented, which is able to withstand disturbances and oscillations and finally maintain the stable performance of the system. In order to investigate and analyze the proposed control method, the system studied in this research is simulated with different scenarios such as disturbances and oscillations in Simulink environment of MATLAB software. Evaluation of the simulation results proves the accuracy of the performance and the efficiency of the proposed controller. The proposed strategy by applying inertia and virtual damping, acceptably reduces the voltage oscillations of the DC microgrid and finally improves the stability of DC microgrid.

The stock market in general is very unpredictable in nature. Many factors may play a role in determining the price of a particular stock, such as market trends, supply and demand ratios, the global economy, public sentiment, sensitive financial information, earnings announcements, historical prices, and many more. The challenge of accurate forecasting But, with the help of new technologies such as data mining and machine learning, we can analyze big data and create an accurate forecasting model that avoids some human errors. In this work, the closing prices of specific stocks are predicted from sample data using a supervised machine learning algorithm. Specifically, a Recurrent Neural Network (RNN) algorithm is used on stock time series data. The predicted closing prices are checked against the actual closing price. In this research, we investigate the problem of stock market forecasting using Recurrent Neural Network (RNN) with short-term memory (LSTM). The purpose of this research is to investigate the feasibility and performance of LSTM in stock market forecasting. We optimize the LSTM model by testing different configurations, for example, the number of neurons in the hidden layers and the number of samples, respectively. We have used historical stock price data collected from the Yahoo Finance database to train our model. Nevertheless, based on the forecasting results of the LSTM model, we used it to predict the stock value in the coming days for the final stock price. The results show that the LSTM model with 4 layers has a higher accuracy and the lowest error (about 0.1771, 2101 and 0.1617) compared to the rest of the layers for prediction.   

   Abstract Today, the expansion of the use of the Internet has made it possible for millions of users around the world to access online social networks. On the other hand, these networks have been in the focus of users' attention in the dissemination of information, especially in areas such as viral marketing advertising, improving recommender systems, transferring time-sensitive information, guiding public opinion, promoting national security, sociology, etc. One of the important issues surrounding the dissemination of information in online social networks is the issue of effective dissemination of the message at a suitable speed. For this purpose, it is necessary to identify the influential users in a suitable way. In this thesis, it is proposed to identify influential people based on the reactive information of users and according to their sphere of influence. For this purpose, first, the value of network communication is determined based on the reactive information of users (reply and retweet), and then, the structure of the network is divided into its constituent communities. In the next step, centrality criteria are used to evaluate the importance of each member of the community, and at the end, influential nodes are identified in their sphere of influence. The effectiveness of the proposed method using a real database that includes the retweet and reply information of Twitter users; has been tested. Due to the lack of a database that has the tagging of influential people and reactive information at the same time; from one of the methods of information dissemination in social networks that has an acceptable correlation with the proposed method; Used. At the end, the obtained results are compared with previous similar methods. The results of the evaluations showed that the method proposed in this thesis can identify effective users with more accuracy, efficiency and speed. Also, the evaluation results showed that the proposed method can achieve 87.33% detection accuracy in identifying effective users, which shows an improvement of at least 13% compared to the compared methods.    Keywords: social network analysis, identification of influential users, identification of communities, graph analysis.   

  Long-ShortTerm Memory neural networks have high computational complexity, resulting inlong execution times. Hardwareimplementation is one of the proposed solutions to this challenge. However, thepower, delay, and area are serious challenges in the hardware design process. Asuitable solution would be to replace the approximate circuits with exactcircuits. The purpose of this study is to evaluate the efficiency of long-shortterm memory networks using approximate computing units. However, because themultiplication operator is used extensively in the network structure, we turnedour attention to replacing the multipliers. For this purpose, we replaced allthe multipliers in the EvoApprox8b approximate library instead of the exactmultiplier in the network structure for forecasting stock market signals in thedatasets of Apple, Microsoft, and IBM companies, and we examined theperformance of the network. From the simulation results, it was found thatreplacing the approximate multiplier can cause a decrease of 205.8 µw in power,0.64 ns in delay, 236 µm2 in area, and 366.5 J in PDP, to predictthe close signal in the Apple stock market dataset. The substitution effect inpredicting the High signal in the Microsoft stock market dataset was in theform of a decrease of 170 µm2 in area, 38.8 µw in power and anincrease of 17.7 J in PDP, and the substitution effect in predicting the Lowsignal for IBM was in the form of a decrease of 47.1 µw in power, 432.646 J wasobserved in PDP and 87 µm2 in area. In the second part of study,with the help of the inherent hardware criteria of the approximate multipliers,we designed a predictive model to predict the suitability of the multiplier forreplacement in the network hardware structure. Therefore, a binaryclassification problem was defined. Next, using feature-selection algorithms,we determined the number of entries in the desired model. Our proposed model isa Linear Discriminant classifier, which can predict the performance of anapproximate multiplier from the EvoApprox8b library in the LSTM networkhardware structure using the Mean_AED, Correct, and Var_ED inherent errorcriteria with 99% accuracy.

Abstract Due to the progress of science and technology in various cultural, social and economic fields, the need to receive data from various databases based on extracting information patterns to achieve life-giving research results is increasing day by day. On the other hand, maintaining the security of private data of individuals and organizations is a key issue in this field, ignoring it will lead to incorrect information and customer dissatisfaction, and will ultimately lead to incorrect research results. One of the most important information data in this regard is medical data, which even according to medical regulations, respecting people's privacy and keeping information confidential is essential. With the world entering the channel of the global epidemic of the corona virus, controlling the epidemic in the first place and finding its cure in the second place has become the challenge of the scientists and doctors of the world. In this regard, the computer science community has offered its role to control the corona virus epidemic to the world. Using deep learning to detect covid-19 from X-ray images becomes a fast method to diagnose patients and manage care services for patients. To achieve better results, we need a lot of data from different information sources, and data privacy as a barrier in this way will prevent engineers from achieving this important goal. Therefore, by introducing federated learning, which is a nascent leap towards creativity and better results, we will introduce its advantages and challenges and simulate it in the diagnosis of Covid-19 and try to take a small step to achieve more accurate results with more data. And of course, more organized to train neural network models, in the meantime, by presenting a strong aggregate approach, we were able to surpass our competitors with an accuracy of 97.04.      Keywords: Federated learning, covid19, x-ray, deep learning

   Individual differences originate from one's personality, which is the most crucial factor affecting one's decisions and choices in life. In recent years, automatic personality detection from text has attracted the attention of most researchers due to its applications like recognition of qualified managers, job selection, selection of academic courses, and online businesses. However, most current methodologies focus on the statistical features of text and ignore the semantic relations and information. These writings and texts are a way to translate internal thinking and feeling in a manner that is understandable to others. The research aims to automatically predict personality from text using language modeling and deep learning. This research presents a deep learning algorithm, Convolutional Neural Networks, through two approaches for personality prediction from two benchmark datasets: Essay and MyPersonality. In the first approach, whole text is utilized for modeling, training and evaluation; However, in the second approach, key phrases derived by a multipartite graph are used. In order to extract features, three techniques of SentenceTransformer, Longformer and short-time Fourier Transform are presented and applied for the first time in the personality prediction research.

  With the size of transistors reaching the nanometer scale, a large number of processing units (PEs) can be embedded in one chip. With the increase in the number of PE in a chip, the existence of a scalable and powerful communication infrastructure for communication between these cores is necessary. Network-on-chip (NoC) is proposed to meet this need. However, like any other electronic component, the network-on-chip is prone to transient and permanent faults. Among the transient and permanent faults, the transient faults have a higher rate. These faults can reduce the performance of the network-on-chip, and if the rate of faults increases, the whole system is susceptible to failure. Most of the past solutions in the field of fault-tolerance network-on-chip, which have worked in the field of router design, have investigated the problem of single-bit faults. This is despite the fact that due to the shrinking size of transistors, the probability of multi-bit transient faults has increased. Therefore, in this thesis, we have used the Flexible Unequal Error Control (FUEC) coding method in our proposed router to correct single-bit, two-bit, and three-bit faults in the channel and the router's input buffers. We have also used the Triple Module Redundancy (TMR) technique to deal with the problems of router control units. Also, to correct the faults on the multiplexers of cro  ar, we have used redundant information and redundant time to discover and correct the faults, respectively. However, all fault tolerance techniques are associated with hardware overheads. These overheads increase power consumption, while power consumption is an important challenge in today's world. Therefore, it is very necessary to have mechanisms that are power-aware to reduce power consumption. Therefore, in this thesis, a power-aware mechanism has been proposed, which reduces the energy consumption by turning off the idle input buffers in the routers. To evaluate the proposed router, we have implemented it in the Noxim simulator and in this simulator, we have randomly injected faults for synthetic traffic and 8x8 two-dimensional mesh. On the other hand, using the SystemVerilog language, we have synthesized the hardware design of the router in the Vivado2019.1 and obtained the hardware overheads and power consumption using this tool. Using this router in the on-chip network can increase fault tolerance by 5 times and reduce energy consumption by 12% compared to the basic router. On the other hand, this router requires 57% more hardware overhead than the basic router. It also increases the average network delay up to 2 times.

Network traffic classification has been studied for two decades and has been applied to a wide variety of applications, including network traffic management, security in firewalls, and intrusion detection systems. Traditional network traffic classification methods, including port-based methods, deep packet i  ection, and traditional machine learning methods, have been widely used in the past. But due to the dramatic changes that happened in the field of traffic on the Internet, especially the increase in encrypted traffic, as well as the need for these methods to extract features from data streams manually by experts in this field, which was time-consuming, expensive and error-prone. , the accuracy of these methods decreased dramatically. This caused the emergence of newer methods in the field of network traffic classification. Deep learning methods, which are a subset of machine learning science, were able to quickly open their place in this field by automatically extracting features from traffic flows and removing the need for feature extraction by experts, as well as the high accuracy they showed in traffic classification. Also, techniques such as data fusion techniques, as helpful techniques that can be used to further increase accuracy and improve network traffic classification, have come to the aid of these methods. In this research, an attempt has been made to use recurrent deep networks and cumulative cryptography to extract features from the high-level ISCX VPN-non VPN traffic data set. Then, by applying the data integration technique at the feature level, the features extracted from the mentioned networks will be brought to an optimal set of features and finally increase the accuracy in traffic classification. It should be noted that traffic classification is done by the multi-layer perceptron deep network. According to the evaluations, the accuracy of the proposed model in network traffic classification has reached 99.1%.  

   ABSTRACT Audio conversion aims to change one or more aspects of the speech signal while preserving the speech structure of the signal. One of the subcategories of voice conversion is voice conversion. Voice conversion is a technique to transform the identity of the hidden speaker in the source speech waveform while preserving the linguistic information. The goal of the voice conversion system is to create a conversion function, which converts the same speech features of the language from both source and target speakers. By placing the corresponding features of the target speaker with the corresponding features of the source speaker's speech, and reconstructing these features into the speech wave, voice conversion occurs. Most of the topics of voice conversion revolve around learning the corresponding characteristics of the source and target speakers. In this research, it has been tried to convert the speech wave of the source speaker by separating the signal of the source speaker and the target into the same time segments and convert it into a two-dimensional Mel Spectrum matrix (using the MelGAN vocoder), it prepare the input data and train the network Created, i  ired by Cycle GAN, this transformation function. The MelGAN vocoder has been used to synthesize (transform) the waveform into Mel Spectrum and vice versa (speech waveform). Also, in this research, the data of the voice imitation challenge of 2018 was used. The challenge [50], held every two years, attempts to improve the quality of voice imitation by providing data (in the 2018 series, non-parallel data). In the end, two subjective and objective (realistic) methods have been used to evaluate the final transformation function trained in this research. Existing objective evaluation criteria for voice conversion (VC) are not always relevant to human perception. Therefore, training VC models with such metrics may not effectively improve the naturalness and similarity of the converted speech. In this project, evaluation models based on deep learning have been used to predict human ratings from transformed speech. We adopt convolutional and recurrent neural network models to develop a mean opinion score (MOS) predictor, called MOSnet. Also, the MCD criterion has been used for objective evaluation. Despite years of research, voice imitation systems, and the progress of the transformation function learning process, using different types of neural networks, still have deficiencies in accurately imitating a target speaker spectrally and prosodically and at the same time maintaining speech quality.

ا رشد پلتفرم ها و برنامه هاي كاربردي شبكه هاي اجتماعي آنالين، روزانه مقادير زيادي محتواي متني توسط كاربر به روشهاي مختلف مانند نظرات، تحليلها، اخبار و پيام هاي متني كوتاه ايجاد مي شود. در نتيجه، كاربران اغلب براي استخراج اطالعات مفيد در مورد موضوع مورد بحث اين گونه محتوا را پالش برانگيز ميدانند. امروزه براي 1 استخراج راحتتر اطالعات مفيد از روشي به نام استخراج موضوع استفاده ميكنند. استخراج موضوع با اسستقاده از يك سري محاسبات آماري خالصه يا موضوع اصلي سند مورد نظر را از متن بيرون ميكشد، كه با اين كار ميتوان با مشكالت كمتري به تجزيه و تحليل اسناد پرداخت. در اين پژوهش قصد داريم با استفاده از روشهاي يادگيري عميق همچون)DNN,LSTM )يك شبكه يادگيري عميق جهت استخراج موضوع با دقت بيشتر از كارهاي انجام شده در اين زمينه طراحي كنيم. ديتابيسي كه در اين پژوهش بر روي آن كار خواهيم كرد ديتابيسي متني شامل اخبار است. كه در ابتدا با استفاده از تكنيكهاي پيشپردازش متن )تبديل كردن تمامي حروف موجود در دادههاي متني به »حروف كوچك« )letters Lowercase ،)پاك كردن عالئم نقطهگذاري )Punctuations ،)پاك كردن »كلمات بي اثر« )Stopwords ،)مصدر سازي كلمات)Stemming ) )عمليات نرمل 2 سازي را انجان داديم. از LDA بعنوان روش يادگيري شبكه استفاده ميكنيم به بياني واضح تر شبكه يادگيري عميق بر اساس تكنيك استخراج موضوع LDA كار خواهد كرد. نتيجه اين پژوهش دقت بشتر شبكه يادگيري نسبت به شبكههاي ساخته شده در كارهاي پيشين است كه توانستهايم دقت شبكه بر روي ديتابيس مورد نظر را نسبت به آنها بيشتر كنيم.  

  Convolutional neural network has been used in the cuckoo filter for the named data network. First, this cuckoo filter has been made two-dimensional, then a neural network has been used for training. The purpose of this training and learning method is to extract the features of the inserted data and use those features during the data search, which ultimately improves the search speed.

Abstract

A wireless sensor network consists of sensor nodes located in geographic areas whose job is to monitorPhenomena such as humidity, temperature, vibration, etc. Since wireless sensor networks are networks that are used to monitor and control the environment, data transfer in these networks is based on data. Is done and routing protocols in these networks must be data based.

بدون شك امروزه يكي از چالش ­هاي مهم مراكز داده ابري مصرف انرژي بسيار زياد است. از طرفي احتمال خرابي يك سرور در يك مركز داده با تعداد زياد سرور و از دست رفتن ماشين ­هاي مجازي روي آن امري اجتناب ناپذير است. يك روش سنتي براي افزايش قابليت دسترسي سرورها و در نتيجه افزايش قابليت اطمينان ماشين مجازي استفاده از افزونگي مي­ باشد بطوريكه تعداد ماشين مجازي يكسان روي سرورهاي مختلف اجرا شوند كه در صورت بروز خرابي در يك سرور نسخه ­هاي پشتيبان روي سرورهاي ديگر كار مورد نظر را انجام دهند. استفاده از افزونگي منجر به افزايش مصرف انرژي در مراكز داده ابري مي­ شود، لذا بهينه ­سازي اين دو پارامتر نياز به مصالحه دارد. اين پايان نامه روشي جهت تخصيص ماشين ­هاي مجازي با در نظر گرفتن مصرف انرژي و قابليت اطمينان را تحت عنوان روش P21(Placement method with 2 active replication and 1 inactive replication) ارائه مي ­دهد. ايده اصلي روش پيشنهادي در نظر گرفتن دو نسخه­ ي فعال و يك نسخه­ ي غيرفعال افزونگي مي ­باشد. جايگذاري ماشين­ هاي مجازي روي سرورهايي با بالاترين مقدار تابع هدف صورت مي ­گيرد. در همين راستا روش ارائه شده، 2 نسخه­ ي فعال از افزونگي را به عنوان نسخه ­ي اصلي و نسخه­ ي اول پشتيبان در نظر مي­ گيرد و يك نسخه­ ي غير فعال افزونگي را به صورت رزرو دارد. در ابتداي كار، زماني كه ماشين مجازي بر روي سرور نسخه ­ي اصلي جايگذاري مي­ شود همزمان يك نسخه­ ي پشتيبان به صورت سينك روي نسخه­ ي پشتيبان فعال پردازش مي­ شود. زماني كه سرور دچار مشكل شود به دليل حفظ دسترس ­پذيري، عمليات فعال سازي سرور پشتيبان غير فعال آغاز مي­ گردد. بلافاصله، از اطلاعات و پردازش ­هاي صورت گرفته بر روي سرور نسخه­ ي فعال يك نسخه (image) تهيه مي­ شود و پس از فعال سازي سرور غير فعال به آن منتقل مي­ گردد. به همين روش مي­ توان قابليت اطمينان سيستم را بالا برد و با استفاده از قابليت رزرو بودن نسخه­ ي پشتيبان مي­توان از تعداد سرور كمتري استفاده كرد. به دليل عدم پشتيباني قابليت اطمينان در شبيه ­ساز كلودسيم، جهت ارزيابي روش پيشنهادي يك شبيه ­ساز به زبان جاوا پياده ­سازي شده ­است. در اين ارزيابي 8 آزمايش با باركاري مختلف از جمله زمان ورود و پايان تسك­ ها، تعداد هسته­ هاي مورد نياز هر تسك و تاخير بين تسك­ ها مورد بررسي قرار گرفته­ است. اين 8 آزمايش به ازاي تفاوت در تعداد سرورها، توان، دسترس پذيري و تعداد هسته مي­ باشند. نتايج شبيه ­سازي نشان مي­ دهد كه مصرف انرژي براي حالت­ هاي مختلف بين 38-1 درصد كمتر و قابليت اطمينان بين 58/2-01/0 درصد افزايش را نسبت به روش­ هاي مشابه دارد.كلمات كليدي: رايانش ابري، جايگذاري ماشين ­هاي مجازي، قابليت اطمينان، مصرف انرژي، افزونگي

  Abstract Images are the most important and widely used digital data used in computer systems. A digital image is made up of a set of objects or areas, so one of the efficient techniques for extracting features from images with respect to their constituent objects is the image segmentation technique, which delimits objects or areas. Highlights the image with high accuracy due to its texture and features. Using image segmentation, image pixels are placed next to each other in specific areas due to common features and generally similarity to each other. Multi-level image thresholding is one of the most popular and at the same time the simplest and most efficient methods of image segmentation. The most important issue in this method is the selection of the value of the relevant thresholds. In such a way that by determining the appropriate thresholds, the desired image can be more accurately zoned. Atsu method is one of the thresholding methods that has a good performance in determining two-level thresholds, but when increasing the number of thresholds, Atsu performance decreases in terms of time and segmentation accuracy. Therefore, it is combined with optimization algorithms to achieve better performance in terms of time and segmentation accuracy. In this research, an improved Grasshopper optimization algorithm is also proposed to increase the accuracy of finding answers and increasing the accuracy of segmentation, as well as to increase image quality. In this method, Atsu evaluation function is proposed for the image segmentation process in optimization algorithms. According to experiments and results, the improved Grasshopper algorithm is performs better compared to the optimization algorithms for Grasshopper, whale, firefly and bee colony. Keywords: Image segmentation, Improved Grasshopper Optimization Algorithm, Grasshopper Optimization Algorithm, Firefly optimization algorithm, Artificial Bee colony algorithm and Whale optimization algorithm. Abstract Images are the most important and widely used digital data used in computer systems. A digital image is made up of a set of objects or areas, so one of the efficient techniques for extracting features from images with respect to their constituent objects is the image segmentation technique, which delimits objects or areas. Highlights the image with high accuracy due to its texture and features. Using image segmentation, image pixels are placed next to each other in specific areas due to common features and generally similarity to each other. Multi-level image thresholding is one of the most popular and at the same time the simplest and most efficient methods of image segmentation. The most important issue in this method is the selection of the value of the relevant thresholds. In such a way that by determining the appropriate thresholds, the desired image can be more accurately zoned. Atsu method is one of the thresholding methods that has a good performance in determining two-level thresholds, but when increasing the number of thresholds, Atsu performance decreases in terms of time and segmentation accuracy. Therefore, it is combined with optimization algorithms to achieve better performance in terms of time and segmentation accuracy. In this research, an improved Grasshopper optimization algorithm is also proposed to increase the accuracy of finding answers and increasing the accuracy of segmentation, as well as to increase image quality. In this method, Atsu evaluation function is proposed for the image segmentation process in optimization algorithms. According to experiments and results, the improved Grasshopper algorithm is performs better compared to the optimization algorithms for Grasshopper, whale, firefly and bee colony. Keywords:

Approximatecomputing   are a promising technique for reducingpower consumption or improving circuit delay, with the help of which a suitable trade-offcan be achieved between power consumption, delay, and accuracy in circuitoutput. In this work, we propose approximate multiplication circuits withdifferent bit widths and the effect of using 8-bit multiplication in imageprocessing algorithms such as: Gaussian filter smoothing algorithm, ContrastStretching algorithm, edge detection algorithm with sobel-filter, andmultiplication of images algorithm. The proposed approximate base multiplier isa 4-bit multiplier that divides the circuit into two parts to reduce circuitdelay, the lower part is free of carry, and the upper part has a 2-bit carrychain independent of the lower part. To expand the circuit and produce 8-, 16-,and 32-bit multiplier circuits, we use a 4-bit base multiplier, and we useseven techniques for final accumulation and summation. The adder used for thefinal summation is an accurate adder and an approximate adder available withdifferent configurations to have accuracy at different levels. The results of approximate multiplier implementation show that theproposed 4-bit multiplier has a maximum of 11.55%, 11.75%, 7.99%, 45.64%,53.21%, 68.57%, 82.91% and 94.63% improvement in parameters Mean Error Distance(MED), Mean Normalized Error Distance (NMED), Mean Relative Error Distance(MRED), Power Consumption, Area, Delay, Power-Delay Product (PDP), and Energy-DelayProduct (EDP), respectively, compared to existing 4-bitmultiplication. In the proposed 4-bit multiplier, 72.81%, 74.35%, 83.33%, 95.46%and 99.24% improvement in parameters Power Consumption,Area,Delay, Power-Delay Product (PDP), and Energy-Delay Product (EDP), respectively, compared to accurate wallacetree multiplier. The results of using 8-bit multipliersin the mentioned image processing algorithms also show the acceptable qualityof the processed images.

The Internet of Things is a network of physical objects connected by the Internet. The Internet of Things covers a variety of areas, including home automation, industrial processes, human health monitoring, and environmental monitoring. The future of objects is the future of the Internet and will be useful for anything available in the world. Internet of Things, despite many benefits, also create security and privacy challenges. IoT systems are very vulnerable, so an intrusion detection system requires IoT environments. Intrusion detection systems are an important tool for protecting networks and information systems. The purpose of an intrusion detection system is not preventing attack and only discover and identify attacks and identify security bugs in the system or computer network and its announcement to the system administrator. Despite the fact that decades are developed from the development of previous influence systems, these systems still face challenges to improve diagnosis accuracy. Many intrusion detection systems are still suffering from high false alert rates, so many researchers have focused on developing intrusion detection systems with high detection rates and reducing the wrong alert rate. Since the network environment changes quickly, a variety of new attacks appear. Therefore, we need to develop intrusion detection systems that can identify unknown attacks. To solve these problems, researchers have begun to focus on building intrusion detection systems using machine learning methods. Deep learning is a branch of machine learning, based on a set of algorithms that are trying to model high-level abstract concepts in the data. Therefore, in this dissertation, the process of detecting intrusion on the Internet of things with a deep learning approach has been addressed. In this thesis, an attack identification method and anomalies based on the combination of deep learning CNN-LSTM algorithms are used in the Bot-IoT data set. In order to evaluate performance, the indicators of accuracy, accuracy and reminders have been used. According to the results obtained in the proposed method, the detection accuracy of 99.98% is obtained.   

In the last century, remote sensing imagery has been a major source of information in many applications, such as land cover detection, resource management, and monitoring. These images include a variety of aerial and satellite imagery, the use of which has expanded dramatically by placing various cameras and sensors on airplanes and other aircrafts. Detection and extraction of surface water is one of the main applications of remote sensing images that play a key role in controlling resources and preventing floods and crises such as drought. To date, various methods such as image thresholding, index detection, edge-based detection, and machine learning methods such as support vector machine have been used to improve the quality of water detection in images; However, the main application of these methods has been in problems where the water areas are not very dispersed and the water body has smoother boundaries and in images that include challenges such as the existence of water dispersed areas or narrow rivers, almost none - did not provide acceptable accuracy. Despite these issues, deep neural networks have obtained the state of the art results in the field of remote sensing image segmentation. In this research, a hybrid architecture called "stacked ensemble model" is presented to pixelwise >Based on the obtained results, the stacked ensemble method proposed in this thesis has succeeded in receiving the best result and also achieving the first rank among the participants of AIcrowd LNDST water body segmentation challenge which was held in August 2020.[1]    Key Words: Remote Sensing, Satellite Imagery, Water Body, Surface Water, Semantic Segmentation, Deep Learning [1] https://www.aicrowd.com/challenges/ai-for-good-ai-blitz-3/problems/lndst/leaderboards   

   One of the most common types of seismic isolators is steel reinforced elastomeric isolators (SREIs) which consist of alternating layers of elastomer and steel reinforcing plates. Unbonded fiber-reinforced elastomeric isolators (UFREI) are a relatively new type of elastomeric isolators. In this type of isolator to control lateral strain and provide vertical stiffness, FRP layers are used instead of steel plates. Additionally, to reduce the cost of isolators, the idea of removing the top and bottom connection plates and the unbonded use of isolators has been considered. In UFREIs, due to the rollover deformation and the reduction of the isolator horizontal stiffness, it is expected that the seismic isolation efficiency is increased as compared to the bonded isolators. In this research, the performance of UFREIs and conventional SREIs in improving the seismic behavior of liquid storage tanks were evaluated and compared. The isolated water tank was modeled using a mass and spring model of three degrees of freedom with convective mass, impulsive mass, and rigid mass. Time history analyses were performed on the fixed-base storage tank, as the benchmark structure, and the two base-isolated tanks with steel-reinforced and unbonded fiber-reinforced isolators. The results show that both types of isolators are effective in significantly decreasing the demand base shear in the tanks. However, seismic isolation increases the displacement demand in the convective mass. Regarding the comparison of the two types of isolators, it was observed that on average, UFREIs in slender and broad tanks are respectively 33.5% and 23.9% more efficient than the SREIs in reducing the maximum base shear forces. Also, there is no significant difference in the maximum displacement of the convective mass in the two isolation systems. The displacement and shear forces developed in the unbonded isolators were found to be less sensitive to the variations of the peak ground acceleration (PGA) as compared with the conventional bonded isolators.

طبقه­بندي كاني­ها بخش جدايي­ناپذيري از زمين­شناسي است. به‌صورت سنتي براي مطالعه­ كاني­هاي موجود در مقاطع نازك، مرز بين كاني­ها به‌صورت دستي جداشده، هر ناحيه برچسب­گذاري و درصد هر كاني محاسبه مي­شود. اين روش نيازمند دانش، تخصص و تجربه­ بالايي است. از سوي ديگر خطاي انساني ناشي از خستگي و بي­دقتي موجب كاهش دقت طبقه­بندي مي­شود. بنابراين به­كارگيري يك سامانه مبتني بر پردازش تصوير براي تشخيص خودكاركاني­هاي موجود در سنگ­ها امري ضروري است. ارائه چنين سامانه­اي مي­تواند باعث افزايش دقت، كاهش خطاهاي انساني، كاهش هزينه و كاهش زمان جهت تشخيص نوع كاني­ها مي­شود؛ بنابراين، هدف اين پژوهش، پيشنهاد يك سامانه تشخيص خودكار كاني است كه با استفاده از پردازش تصوير، كاني­هاي موجود در سنگ را شناسايي و طبقه­بندي كند. مرحله اول در انجام اين پژوهش ايجاد يك پايگاه داده از تصاوير مقاطع نازك سنگ است. اين مرحله يكي از چالش­برانگيزترين مراحل اين پژوهش بود، زيرا ايجاد يك پايگاه داده مناسب از تصاوير مقاطع نازك، فرآيندي سخت و وقت­گير است. از سوي ديگر، پايگاه داده مشترك و استانداردي در اين حوزه وجود ندارد و هر پژوهشي از پايگاه داده متفاوتي استفاده مي­كند. پس از ايجاد پايگاه داده و برچسب­گذاري تصاوير مقاطع نازك، چند روش­ قطعه­بندي بررسي و الگوريتم JSEG براي قطعه­بندي انتخاب شده است. پس از انجام قطعه­بندي، ويژگي­هاي مبتني بر رنگ و بافت از هر ناحيه استخراج شده­اند. ويژگي­هاي رنگي از هر دو فضاي رنگي RGB و HSI استخراج شده­اند. هم­چنين به دليل اينكه برخي كاني­هاي متفاوت داراي رنگ­هاي مشابه هستند، ويژگي­هاي بافت نيز از هر ناحيه استخراج شده­اند. ويژگي­هاي استخراج‌شده از هر ناحيه، براي طبقه­بندي به طبقه­بند فرستاده شده و طبقه­بند هر ناحيه را به‌عنوان يك كاني برچسب­گذاري كرده است. در اين پژوهش كارايي شش طبقه­بند Linear Discriminant، Su  ace Discriminant، Boosted Tree، Bagged Tree، Linear SVM و Weighted KNN بر اساس معيارهاي مختلف مورد ارزيابي قرار گرفته است. بر اساس نتايج تجربي به‌دست‌آمده، طبقه­بند Bagged Tree داراي بالاترين دقت به ميزان 5253/95 و همچنين كمترين ميزان خطاي MAE برابر با 0447/0 و خطاي RMSE برابر با 2115/0 مي­باشد. همچنين همه طبقه­بندها داراي دقت قابل قبول بالاي 93% هستند. اين نتايج نشان مي­دهد كه روش پيشنهادي داراي قابليت مناسبي جهت شناسايي خودكار كاني­هاست.  

پيشرفت فنّاوري­هاي ساخت تراشه­هاي سيليكوني و دستيابي به ابعاد نانومتري، امكان ساخت سيستم­هاي الكترونيكي بزرگ بر روي يك تراشه را فراهم نموده است. اين تراشه­هاي جديد با چالش­هاي جديدي نيز مواجه هستند و ممكن است در هر زماني در محيط كار دچار اشكال شوند. اين امر نياز به روش­هاي خودآزمايي دوره­اي در محيط كار را بيشتر كرده است. استفاده از روش­هاي خودآزمايي سخت­افزاري به علت ويژگي­هاي آزمون تصادفي نمي­تواند به تنهايي كافي باشد و براي رسيدن به سطح كيفيت مناسب براي آزمون پردازنده، بايستي از روش­هاي خودآزمايي نرم­افزاري نيز بهره برد. يكي از مهمترين مراحل در فرآيند آزمون يك تراشه، توليد بردارهاي آزمون كارا براي آزمون آن تراشه است. توليد بردارهاي آزمون با استفاد از روش­هاي قطعي توليد آزمون بسيار زمانبر است. علاوه بر اين محدوديت­هاي زماني و عملكردي در پردازنده­ها، باعث مي­شود كه توليد آزمون­هاي نرم­افزاري براي پردازنده با استفاده از روش­هاي قطعي توليد بردار آزمون، ناممكن و يا حداقل سخت و ناكارا باشد. استفاده از روش­هاي توليد آزمون مبتني بر شبيه­سازي به دليل غلبه بر اين محدوديت­ها، مي تواند يك جايگزين مناسب باشد. در روش­هاي مبتني بر شبيه­سازي، تعدادي الگوي آزمون به صورت كاملاً تصادفي و يا با استفاده از روش­هاي فرا ابتكاري توليد مي­شود. سپس اين بردارهاي آزمون بر اساس شاخص پوشش شكال، مقايسه شده و بهترين آنها انتخاب مي­شوند. در اين روش­ها، محاسبه­ي شاخص پوشش اشكال بردار آزمون زمانبر است. مي­توان به­جاي شاخص دقيق و زمانبر پوشش اشكال، از يك شاخص تقريبي و سريع براي ارزيابي و انتخاب بردارهاي آزمون استفاده نمود.   در اين راستا، در اين پايان­نامه يك شاخص تقريبي به نام APXD پيشنهاد شده است كه تقريبي مناسب از تعداد اشكال­هاي شناسايي شده توسط يك الگوي آزمون ارائه مي­كند. با تكيه­ بر اين شاخص، يك روش توليد آزمون مبتني بر شبيه­سازي به نام APXD_TG نيز در اين پايان­نامه پيشنهاد شده است و با استفاده از آن براي برخي از اجزاي يك پردازنده، آزمون نرم­افزاري توليد شده است. نتايج ارزيابي­هاي ما نشان مي­دهد كه شاخص APXD سرعت و دقت مناسب داشته و جايگزين مناسبي براي شاخص پوشش اشكال است. علاوه بر اين نتايج ارزيابي­ها نشان مي­دهند كه استفاده از شاخص APXD به جاي شاخص پوشش اشكال، ضمن حفظ كيفيت آزمون، زمان توليد آزمون را به نحو قابل توجهي كاهش مي­دهد. شاخص پيشنهادي بسيار سريع­تر از شاخص پوشش است. به طور ميانگين نسبت به روش موازي ?? برابر سريع­تر و نسبت به روش سريال 696.9 برابر سريع­تر است. لذا استفاده از آن در بخش ارزيابي بردارهاي آزمون كانديد، تسريع قابل توجهي در الگوريتم‌هاي توليد آزمون مبتني بر شبيه­سازي به وجود مي­آورد.   

different places (e.g. emergency rooms, bank branches, etc.) is long queues which cause them to waste their time. By designing a system that can provide people with information about the state of congestion in different places, if it used by people, it can help them to avoid wasting their time. This system can be very useful when citizens are asked not to be present in crowded places due to the spread of infectious diseases such as coronavirus. In this thesis, an IoT system based on edge computing architecture is designed to solve the issue of queue formation and population overcrowding. In this system, a Windows-based software receives images by connecting to wireless CCTV cameras and counts the number of people with the help of an image processing method, and then, sends the location status to the server. In this thesis, the background, studies conducted, and challenges of IoT-based people counting systems and image processing algorithms are discussed, and three image processing methods for counting people are proposed. Two methods are based on eliminating the background and counting people based on the background pixels, and the third method is a model based on the MRCNN networks, which is taught to count people’s heads. The places that this system is implemented to detect congestion in them are indoor places where there is rarely the problem of people’s shadows. The mean errors of the best method based on the MAE and RMSE scales in all the test frames obtained as 1.51 and 1.89, respectively.

  The main purpose of human data to collection is to understand the thinking of other human beings. This unconscious tendency has led researchers to analyze information in order to understand and analyze the minds of other human beings. Today, with the advancement of information platforms such as the Internet, social networks, etc., it is easy to gather the information you need. Today, social networks are one of the most important aspects of people's lives, and on the other hand, these networks have made huge profits by exploring the general information of users. The aim of this study is to investigate the text to find out the mood of people in typing texts. In this study, 14,000 tweets related to airlines were used to analyze emotions in three categories: positive, negative and neutral. The final proposal has three steps. In the first step, we perform the pre-processing operation on the database. In the second step, using the Imperialist Competitive Algorithm, we extract the main words from all the existing words. Keywords are the words that have the most impact on categorization. We then use the convolution neural network to extract more features. In the last step, we perform the classification operation using the multilayer perceptron neural network (MLP). At the end, using the final proposed design, we achieved precision, accuracy and recall of 0.990, 0.983 and 0.875, respectively. The results indicate that the final proposed design is desirable.

Abstract With the advancement of radar technology due to electronic warfare, radar systems are capable of measuring high-precision targets over long ranges as they are increasingly being used. Radar protection is more important than radar itself. In radar design, the transmitter and receiver position the transmitter and receiver in one place, but they separate the sender and receiver to protect the accuracy of accuracy. Passive radar is a kind of radar used to detect and detect targets using an unknown antenna. Passive radar can be combined with array antennas to preserve more radar security. In this project, a phase array antenna is used to design passive radar. When a number of antennas are located at a distance from each other, they have the potential to continue to operate in the system alone, in other words, some of the antenna may disappear or have a problem, but the rest of the antenna can continue to function with less efficiency. Many parameters have a role in the design of passive radar using phase array antennas. Depending on this radar, this radar has a stationary or mobile location and is used in different applications such as: military, aerial, imagery, photography. The most important part of this thesis is the number of elements in the array. In this design, all-directional antennas (omni-directional) are or can be used with better performance and more complex antennas. All antennas must be synchronous (synchronous), one of the major challenges in this design. At the same time the antenna is complex in a geographic area. To synchronize the antennas using global positioning systems (GPS). To achieve the desired phase, a source antenna with zero phase is used for proper shape shaping elements that have a suitable pattern in one direction according to the desired phase. Keywords:  oscillators,  eam  haping,  antenna  array,  avigation,  assive  radar   

In recent years, internet traffic is growing extremely rapidly with the rapid growth of internet users and the emergence of new applications. As a result, the problem of identifying the applications on the network has become a complex task. The detection and classification of flow patterns and applications on network traffic plays an important role in network security and network management. The purpose of classification is to create a link between packet packets with a particular service or application. The problem with most of the methods is to rely on property extraction by experts. It is difficult and time consuming to find desirable features that lead to high accuracy. In general, most of the traffic classification methods are based on extracted features by an expert on computer networks. These features include port number, packet overhead, packet header and extracted statistical features of flow. The main problem of traffic classification is finding suitable features in traffic network. The process of finding suitable features is time consuming and cost and needs a qualified person to identify and extract these features and to solve these problems, one of the most recent fields in machine learning is deep learning that is based on artificial neural networks and that feature extraction is done in a hierarchical and automatic mode. In this situation, extracting the automatic feature from the expert will be eliminated and the possibility of human errors is reduced. In this work, our solution show that this approach is capable to identify encrypted traffic and surpass the accuracy achieved by almost every classical method in this area of research. We have used Deep Belief Networks and Convolutional Neural Network that can accurately identify and classification on ISCX vpn-nonvpn dataset.

  ?_ Abtract The present study was conducted under the title "Data exchange Protocol between Appointment services Based on Health Data Exchange Center (IX Health)" in 1398. the purpose of the research is to exchange information between health systems. the ability to communicate between appointment systems is one of the key factors in patient satisfaction in receiving medical services, reducing patient and physician waiting time, and so on. in this research, we first examine the systems integration architecture as well as the architecture of the National Center for Information Exchange and the National Center for Health Services and then examine the protocol for data interchange between the systems based on the protocol presented in the electronic health record the proposed protocol focuses on the possibility of data interchange between the delivery systems by providing a communication protocol implemented with the use of php programming language, larval framework and phpstorm environment, the results and outputs of the program show that it is possible to exchange data between queuing systems by providing communication protocol. obviously, this reduces the waiting time for the patient and the physician to increase speed and improve efficiency in medical centers. we also showed that using this communication protocol, it was possible to refer from one system to another. Keywords: Data Exchange, Scheduling. Health

Abstract The advancement of technology and the Internet has led to an ever-increasing growth of databases, especially images, which has led to the search for the desired image and its recovery from the massive amount of databases. searching for images from the past has been an important research topic and several methods have been proposed, including methods for image retrieval based on text, the text-based retrieval method is a basic method and performs searches using the keywords defined for each image, given that the method of text search was a time consuming and costly method. attempts toward other methods and techniques, namely, image retrieval based on content, were made using descriptors of structural elements or low-level features of the image, ie, color, texture and shape, so that we can look at the search image. in this research, we have tried to describe the structural elements of SED and compare it with other descriptors and algorithms that are implemented in this implemented project and to achieve a higher degree of accuracy. by researching and investigating methods and descriptors of structural elements that utilize low-level features of color and texture, the proposed combination method is presented using structural elements and color difference histograms. on the other hand, considering that changing the size of images is an important issue and accessing the image with different sizes is considered an important issue, so the results of different methods of extracting features in 128× 128, 64× 64, 32× 32, 16× 16 and 8×

  Deep Packet I  ection has always been a challenge of performance and a matter of throughput in computer networks. Therefor a lot of different methods have been invented to enhance the operation of DPI in networks. Using probabilistic filters for DPI is an approach which has been taken is recent years. Probabilistic filters are some kind of data structure which are used for membership test among a set of items. These filters can result a false positive answer. One of constraints of using probabilistic filters is incapability of efficient scaling specially when they are used in a software running by CPU. To solve this problem implementing DPI utility on a scalable parallel architecture can be a good solution. On the other hand, emergence of new networks paradigms like software defined networks added new difficulties in monitoring networks. In the base situation, to perform deep packet i  ection in a software defined network, the whole task is delegated to the controller and this makes the controller overloaded thus creating a network bottleneck. This situation created an intensive need for an architecture and new design of deep packet i  ection which is fast, scalable and flexible to fit in SDN networks. The new design should also decrease the workload of controller which is related to deep packet i  ection. In this thesis we try to design, implement, and evaluate a new method that hits needed criteria.

Human Identification Based on Ear Biometric Employing a Hybrid Approach  

  Fundamental frequency estimation is one of the most important issues in the field of speech processing. An accurate estimate of the fundamental frequency plays a key role in the field of speech and music analysis. So far, various methods have been proposed in the time- and frequency-domain. However, the main challenge is the strong noises in speech signals. In this paper, to improve the accuracy of fundamental frequency estimation, we propose a method for optimal combination of fundamental frequency estimation methods, in noisy signals. In this study, to discriminate voiced frames from unvoiced frames in a better way, the Voiced/Unvoiced (V/U) scores of four pitch detection methods are combined both linearly and nonlinearly. These methods are: Autocorrelation, Yin, YAAPT and SWIPE. After identifying the Voiced/Unvoiced label of each frame, the fundamental frequency (F0) of the frame is estimated using the SWIPE method. The optimal coefficients for linear combination are determined using the regularized least squares method with Tikhonov regularization. To evaluate the proposed method, 10 speech files (5 female and 5 male voices) are selected from the PTDB-TUG standard database and the results are presented in terms of SDFPE, MFPE, FPE, GPE, VDE, PTE and FFE standard error criteria. The results of the experiments indicate that the linear combination method (on various SNRs) made GPE error 22.98%, VDE error 26.16%, PTE error 9.26%, and FFE error rate of 32.72% (relative) And the nonlinear combining method reduces the GPE error by 30.64%, the VDE error by 33.58%, the PTE error by 9.58%, and the FFE error by 39.86%, as compared to the popular speech frequency extraction methods. 

  Nowadays different approaches of signal processing are used in many applications due to its potential applicability to a wide range of problems, such as telecommunication and biomedical signals processing. Voice activity detection (VAD) is one of most important signal processing branches in audio signal processor and is used in many telecommunication systems such as Speech compression , speech recognition, upgrade of speech , noise estimation and noise removal. VADs are also used to detect input signals and >For instance in a mobile telecommunication system usually 60% of talk-time includes speech signal, so that the rest of the signal is not informative. To decrease channel capacity and power consumption in this case VADs can be used to submit only pure speech signal. There have been already many studies in this field, however the efficiency of proposed approaches are highly depends on background noise. So that their efficiency may decrease while noise power is quite higher compared to speech power. This current seminar aims to provide an efficient method which is based on the combination of typical detection techniques of the speech versus non-speech blocks, so that the result can be applied for both clean and high SNR environment

  Abstract Early agricultural production of lignocellulose, available at a significant quantity and low cost, can be an additional source of income for farmers without adversely affecting the fertility of the soil for industrial applications. Wheat straw is one of the most abundant agricultural lignocellulosic biomass, partly Organic Wheat Plant. Wheat straw is used after harvest as a feed for livestock and in cattle-breeding buildings, as well as a large amount of it is burned, while it contains high amounts of cellulose and It can be used as an inexpensive early raw material for the production of valuable cellulose derivatives. In this research, the possibility of producing valuable carboxy methyl cellulose material was investigated using cellulose extracted from wheat straw. Wheat straw contains approximately 33-40% cellulose, which together with hemicellulose and lignin make up its main components. In this research, cellulose in wheat straw was first extracted by sodium hydroxide 10% w / w and sodium hypochloride in two stages. The extracted cellulose was converted to Carboxy methyl cellulose using the Williamson ether process, which the FTIR spectrophotometry used to identify the substituted carboxy methyl groups on the cellulose. The degree of substitution (DS) is the most important factor influencing the solubility and application of carboxy methyl cellulose, and its production efficiency is an important factor in the economic process of production, so the etherification process over operating factors, including: a weight ratio of monochloroacetic acid to cellulose, the concentration of sodium hydroxide, the temperature and reaction time were, at three levels (1, 1.4, 1.8), (20, 30, 40% w / w), (30, 50, 70 ° C) and (1, 3, 5 hours) were optimized with design expert software called MODDE by CCF method using response surface methodology in order to achieve the highest DS and an economical efficiency. The optimization results showed that at a weight ratio of monochloroacetic acid to cellulose 1.3 to 1, the concentration of sodium hydroxide was 20% w / w, temperature 70 ° C and reaction time 1 hour, obtained the highest degree of substitution   0.975 and the yield under these conditions was 1.37 g CMC / g cellulose. Keywords: Wheat straw, Cellulose, design expert software, Carboxy methyl cellulose, Degree of substitution   

  Fossil fuels are declining as sources of human consumption over the years, and in addition to the greenhouse gas emissions that result from these fuels, they are damaging our planet. Renewable energies as a solution to reduce the adverse effects of reducing fosil energy reserves, resulting in an intensification of the global energy crisis as well as an increase in greenhouse gas emissions resulting in harm to the environment. In addition to using these resources, network planning should be done in such a way as to minimize the economic cost to the network. In this case, power plants with a higher fuel cost with less power and low fuel-efficient power plants enter the service with more power. This operation is performed according to load balancing equations and pseudo-security assertions, which is known as the optimal load distribution name. If the problem of optimal load transfer includes random input variables such as the power output of the wind power plant, the output of the problem will also be random, so this type of problem-solving optimal load transfer problem is called optimal load probability. In this paper, for the first time, the Cultural Algorithm (CA) is applied to solve the probable optimal power flow problem in the presence of wind power plants uncertainties and taking into account the power network constraints. The Cultural Algorithm is used as a general optimization method for nonlinear and non-convex functions and can substantially shift the target functions with respect to the problem constraints towards the optimal solutions. In this paper, various methods regarding to generation changes in the structure of the algorithm are also compared and investigated. In addition, the improvement methods of the proposed algorithm applied to limited input problems has been discussed and detailed, especially on optimal power flow problem concluding the uncertainties. The ability of the presented algorithm in optimizing the described problem for IEEE standard 30 and 57 buses test cases has been addressed and challenged using MATLAB software among the other well-known algorithms in this field.

بررسي عملكرد ميراگرهاي جرمي تنظيم شونده ويسكوالاستيك 

معرفي روشي براي تحليل احساس داده هاي حجيم (مطالعه موردي تويتر)

Communication problem with a simple idea can be transferred to a facility management network infrastructure / system / centralized system, solved, so that hardware can remain as part of the network data (like hardware available) and tooling to the control unit to Annie on the device. Hadoop has given birth several years, the question arises whether its functioning can be improved. The answer can be quite overwhelming with the composition and performance of Hadoop-based software and networking replied. with the networking issues, the work will pay its processing and network-based software this task is delegated.

چكيدهبا توجه به افزايش ترافيك و نياز به پاسخ گويي سريع به درخواست­ها دسته­بندي بسته­ها به يك تكنولوژي مهم و يك چالش در عملكرد مسيرياب­ها تبديل شده است، بخصوص در زمان همگام سازي تصميم گيري خود با سرعت تبادل داده­ها اين موضوع بيشتر نمود پيدا مي كند، يعني سرعت جست و جوي فيلدها با سرعت لينك­هاي انتقال برابر باشد و تا زماني كه سرعت شبكه­ها ثابت نشود كار روي دسته­بندي بسته­ها اهميت خود را حفظ مي­كند. افزايش روزافزون داده­هاي انتقالي و پويا بودن آنها باعث شده راه حل­ها و معماري­هاي سخت­افزاري يا نرم­افزاري متعددي براي اين موضوع ارائه شود. الگوريتم­هاي نرم­افزاري با وجود توسعه­پذيري بالايي كه فراهم مي­كنند اما از سرعت پائيني برخوردارند از طرف ديگر راه­حل­هاي سخت­افزاري سرعت خوبي دارند ولي هزينه بالا و قابليت توسعه­پذيري كمي دارند. از اين رو ارائه روشي براي ايجاد مصالحه بين سخت­افزار و نرم­افزار مورد توجه محققان قرار گرفته است. طبقه بندي بسته­ها يك جستجوي چند فيلدي با سرعت لينك ا انجام مي­دهد.در اين تحقيق ، به منظور رفع مشكلاتي كه در بالا ذكر شد از دو فيلتر بلوم و خارج قسمت استفاده شد و به منظور انطباق روش جستجو با بسته هاي ارسالي در تعداد فيلدهاي موجود در معماري نوين SDN، اين تعداد به 15 فيلد سرايند افزايش يافت. در نهايت با استفاده از ابزارهاي در دسترس از جمله   Intel Platform Power Estimation Tool (IPPET)   معيارهاي مورد نظر براي بررسي قابليت هاي روش ارائه شده استفاده گرديد و از نتايج حاصل از دو فيلتر برتري فيلتر بلوم نسبت به فيلتر خارج قسمت دربرخي معيارها اثبات گرديد به اين صورت كه در مورد زمان مصرفي، سرعت انجام الگوريتم، توان عملياتي و انرژي مصرفي فيلتر بلوم عملكرد بهتري داشته ولي در موارد حافظه مصرفي و نرخ خطاي مثبت فيلتر خارج قسمت عملكرد بهتري دارد.

  موضوع بهبود كيفيت صدا امروزه به يكي از موضوعات مهم و اساسي روز تبديل‌شده است .ازاين‌رو بهبود گفتارهاي آغشته به نويز يكي از موضوعات مهم در حوزه پردازش سيگنال است و در موارد بسياري مثل تشخيص صدا، شناسايي احساسات صوتي و...كاربرد دارد. تضعيف نويز به‌نحوي‌كه اختلالي در سيگنال اصلي به وجود نياورد يك چالش مهم براي بهبود صدا محسوب مي‌شود. روش‌هاي مختلفي براي كاهش نويز ارائه‌شده‌اند كه ازجمله روش‌هاي پايه مي‌توان به روش تفريق طيفي ، تبديل موجك، و...ساير موارد اشاره كرد. موضوع تحقيق اين پايان­نامه نيز بررسي نويز موجود در سيگنالِ گفتار، حذف و يا كاهش آن نويز ازسيگنال گفتارِنويزي و ايجاد بهبود در سيگنال‌هاي گفتارِ آغشته به نويز مي­باشد.در اين پايان­نامه دو   روش جديد براي كاهش نويز موجود در سيگنال گفتار نويزي ارائه داده ايم . در روش اول ، يك روش تخمين نويز براي نويزهاي غير ايستان همراه با اعمال تبديل موجك بر روي سيگنال و استفاده از الگوريتم بهينه‌سازي گروه ذرات با رفتار كوانتومي،را به صورت تركيبي با روش Bayesian ارائه داده‌ايم تا نويزهاي موجود در سيگنال نويزي را حذف كند و سيگنال بازيابي شده به سيگنال اصلي نزديك‌تر باشد.در روش دوم نيز با اعمال تبديل موجك بر روي سيگنال و تركيب آن با روش SMPR   روشي جديد براي كاهش نويز ارائه داده ايم. روش­هاي پيشنهادي نسبت به روش‌هاي موردتحقيق در اين پايان­نامه بهتر عمل مي‌كنند و منجر به كاهش نويز از سيگنال با كمترين اعوجاج مي‌شوند.

  DN is a new paradigm in computer networks based on global view provided form separation of data plane and control plane. This separation is possible by means of an API between the switches and the controllers such as OpenFlow. Logical centralized in SDN by global view of the network can help to improve network management, load balancing, routing and security. Logically centralized controller allows SDN load balancer to allocate the new incoming flow to the best possible server, efficiently. SDN load balancers mostly operate on L4 in OSI model and decide based on the L2/4 headers and this conditions cause limitations in implementation of networks when Back-end servers are not replica. In this case a data base is needed to store mapping between content and controller and with each incoming flow to the Frontend load balancer the controller allocates that flow to the server containing the request content. To implement the L7 load balancer (application layer) there are traditional methods such as Delayed Binding and TCP Socket Migration and this project discuss the implementation of Delayed Binding based on SDN concepts and also the best server should be chose regarding to the network global view, traffic load and response time of the Back-end server that contains request content. The implementation of this method is done by using a virtual switch named Open vSwitch in a virtual machine monitor or hypervisor and Floodlight controller and the results of the implementation has been shown in this project. The average improvement of response time in comparison with three other algorithms, the L  RT, Round Robin and Random selection methods are 19.58%, 33.94% and 57.41% respectively. Furthermore, the average improvement of throughput in comparison with three other algorithms are 16.52%, 29.72%, and 58.27%, respectively.

  چكيده با پيشرفت علم و نياز روز افزون به امنيت براي تبائل اطلاعات از كانالي امن بيش از هر چيز ديگري بحث رمزنگاري و استفاده آن در ارسال اطلاعات مورد توجه است. هنر رمزنگاري كه با توسعه و فراگير شدن آن به يك علم تبديل شد شاخه هاي رياضي و علوم كامپيوتر است. از دغدغه هاي مهم مراكز امنيتي مي توان به ارسال تصاوير و محفوظ ماندن ان از هر نوع آسيبي اشاره نمود. ما در روند اين پايان نامه طرح به اشتراكگذاري امن  دو تصوير را بر اساس پروتكل سه طرفه شامير دنبال مي كنيم كه براي اين انتقال امن و رمزنگاري آن ابتدا با استفاده از نگاشت تبديل لجستيك سينوسي به هم ريختگي بين پيكسل ها را ايجاد كرده و سپس با كمك دامنه تبديل  چند پارامتري كسري گسسته زاويه اس به عنوان تابع رمزنگاري انتقال توسط پروتكل سه طرفه شامير انجام مي پذيرئ. و سپس در ا دامه نيز ايده وارد نمودن تسهيم راز را در روند به اشتراك گذاري دخيل نموديم و خواهيم ديد آيا مي توان به اشتراك گذاري ذكر شده را با كمك تسهيم راز نيز انجام داد يا خير؟

  .With the advancement of computer networks and the presence of the Internet,our lifestyles have changed. This also has affected people’s jobs, companies’ andorganizations’ activities. Increasing the communication has led to the need for datasecurity and secure transactions. Information security is critical for communications,especially in financial and economic transactions in the digital world. Cryptographyis used to provide secure transactions. Preventing unauthorized people fromaccessing data is one of the most challenging areas in transmitting information viathe Internet. One common approach for protecting data against adversaries is encryptingit. There are many different encryption techniques. In this work, Ellipticcurve cryptography is used. Elliptic curve cryptography is a public key cryptographymethod, which similar to other cryptography methods such as RSA, provides a givensecurity with a short length key. This thesis presents a method to automate the processof transferring medical information and records while ensuring the security ofthe transaction. Furthermore, a secure method for encrypting data and informationin processes such as authentication, transferring data to the insurance

تعيين نقاط بحراني در شبكه هاي هايبريدي و توسعه مسيرهاي اضافي در شبكه جهت بهبود عملكرد و افزايش ظرفيت و پايداري با ملاحظه به توانايي گره ها

  يكي از مباحث مهم در پردازش صوت، پردازش فايل‌هايي است كه در آن مخلوطي از گفتار انسان، سكوت و موزيك وجود دارد. به عنوان نمونه مي‌توان به فايل‌هاي ضبط شده از رسانه‌هاي راديويي، تلويزيوني و ماهواره‌اي اشاره كرد كه حاوي سيگنال‌هاي صوتي متنوعي هستند.در برخي از كاربردها مانند كاهش حجم، افزايش كيفيت، شناسايي و كاربردهاي ديگر نياز به جداسازي گفتار انسان و يا به عبارتي حذف سكوت، موزيك و يا نويزهاي محيطي از سيگنال‌هاي صوتي به‌وجود مي‌آيد. سيستم‌هاي جداسازي گفتار را مي‌توان نوعي از سيستم‌هاي شناسايي گفتار انسان و يا سيستم‌هاي دسته‌بندي كننده‌ي سيگنال‌هاي صوتي دانست كه از آنها براي جداسازي، شناسايي و يا نشانه گذاري قسمت‌هايي از سيگنال صوتي كه شامل گفتار انسان است، استفاده مي‌شود.براي انجام عمليات جداسازي گفتار انسان از سيگنال‌هاي صوتي از روش‌ها و رويكرد‌هاي گوناگوني بهره گرفته‌مي‌شود. هدف ما در اينجا ارائه روشي مناسب وكارا   در قسمت استخراج ويژگي (feature extraction) و هم‌چنين در قسمت دستبه‌بندي (classification) با استفاده از الگوريتم‌هاي قدرتمند و پيشنهادي و نوين براي رسيدن به دقت بالا و كارايي بيشتر مي‌باشد.

  Energy consumption in Computer networks in recent years, due to the notable grow of the users and demanding of multimedia services have been increased. To preserve the environment, decreasing of energy consumption has been attended, specifically. Energy consumption is investigated from different aspects. In a network, different protocols have been defined which affect on energy consumption. Energy consumption in a protocol is defined based on the generated load on link and necessary time to transfer  the generated load. TCP is a protocol that assures a flow will arrive the destination surely. Therefore, generates a notable volume of the  load because of the  acknowledge  acket  which increase  the load on a related link.In this thesis, energy consumption is investigated  from the software point of view and is tried to decrease the number of acknowledge packets to improve the energy consumption beside of reliability control. The achieved  energy efficiency  improvement in this work is   12.09%. The proposed approach in this work may cause the decreasing of throughput in online networks like VOIP wich can be ignored  generally.

  DNA cryptography is a new branch of cryptography that utilizes DNA as an in- formational and computational carrier with the aid of molecular techniques. Most of the modern encryption algorithms have been broken fully or partially. The world of information security looks for new directions to protect the data and their transmi- tion. The DNA computing in the ?elds of cryptography has been identi?ed as a new hope to create some unbreakable algorithms. In this thesis two digital signatures are discussed and analaysed, the ?rst method uses DNA coding and XOR operation with a symmetric key and the second uses DNA coding, Polymerase Chain Reaction and RSA encryption.

  AbstractThe most significant tool in combating crime is the fight against money laundering. Detecting “suspicious transactions of money laundering” in banks is the biggest challenge of combating money laundering. Lack of attention to the context of the bank account owner’s is resulting in low efficiency of anti-money laundering approaches. The aim of this study is to provide a method of detecting suspicious transactions based on data-mining techniques such as a statistical method for analyzing “contextual outlier transactions” by targeting money launderer’s transactions in integration stage. The method of this study was context analyzing, and its population includes 1.8 million simulated transactions belongs to 1008 people from 48 different contexts over a period of 6 years. Simulators probably distributions came from the Kolmogorov-Smirnov test on 50 person cross-sectional actual bank transactions sample. The transactions collected over field research. Due to the unavailability of sufficient numbers of actual bank transactions, simulated transactions used. By simulating, the ability to create scenarios that may not provide in the real world is possible. Testing the idea of the research resulted in 100% True Positive Rate and 1.14% False Positive Rate, compared to most methods, tangible progress achieved. The study findings showed attending to bank account owner’s context, promotes the quality of the methods used in detecting money laundry.  Keywords: Money-laundry, Context, Contextual Variable, Behavioral Variable, Working Set Window

Abstract- The need to private and sensitive information security on multi-touch devices like smartphones and tablets is one of the main problems in information security. Methods that are commonly used passwords and tokens that have a lot of obstacles and challenges. Biometric authentication methods, these methods are a good alternative to overcome the problems. The introduction of biometric based smartphone touchscreen for user authentication is based on finger touch and movement. The purpose of this Study is to examine method of authentication using biometric behavior based on specific gesture for unlocking the device based in existing designs is safe. In this study, by extracting a large number of features and using Distance learning with Genetic Programming, With high accuracy in authentication based on finger multi-touch touch screen to unlock the device achieved.