Theses supervised by Dr. Öğr. Üyesi Mesut Çevik
23 theses · Altınbaş University
Derin öğrenme algoritmalarına dayalı ağ performansını iyileştirmek için ıot ağlarının eğitimli izlenmesi
The introduction of the IoT has forced the integration of billions of devices in different sectors, thereby creating huge data and change. Nevertheless, IoTs have some few challenges when it comes to internetworking and securing the networks. Networking problems like latency, packet loss, congestion and probable security holes make it imperative that networking headers are monitored and monitored are checked for anomalies. This thesis proposes a deep learning-based approach to real-time IoT network monitoring and anomaly detection, focusing on three models: FFNN, CNN, and MLP are the popular categories of Deep Learning Algorithms. The models were created and built with MATLAB to review IoT network data to identify discrepancies, distinguish malfunctioning nodes, and diagnose future problems. In an effort to enhance the outcome of the models, optimization methods of Adam and Stochastic Gradient Descent with Momentum (SGDM) was used. The models were tested on synthetic IoT data and the results highlighted by using the quality control indicators such as accuracy, precision, recall, and F1-score. However, the results show the proposed methods improve the existing results where MLP and CNN have higher accuracy and anomaly detection rates than FFNN, MLP-93 (92.3%), CNN-94 (94%), DT (78.5%), and SVM (85.7%). This considerable enhancement is due to efficiency of CNN for extracting spatial relationship and MLP to learn non-linear relationship in IoT network data. Comparing with other methodologies implemented in the current research, deep learning models provide a higher level of accuracy in the identification of sophisticated abnormalities, which is beneficial for real-time IoT monitoring. In this case, the CNN model showed remarkable improvement on its capability in the identification of network patterns and the prediction of issues as compared to prior models. These outcomes give a strong signal that deep learning models, including CNN and MLP, are more beneficial for real-time IoT network performance monitoring and anomaly identification than conventional models of machine learning. The work from this research can be generalized as follows to be used in improving the dependability and securability of the IoT networks while giving a much better solution than the current method.
Deep learning-driven classification ofdepth image-based rendering anddynamic face warping in deepfakealtered videos for fabricated news onsocial media
Deepfakes present a persistent challenge in a world driven by transparency and truth. This study introduce a new deepfake detection method based on Xception architecture. Deepfake is having some threats to the authenticity of digital information especially on social media platforms through manipulated content, which transpires fast within these networks and powerfully influences the public's opinion, reputational harm, and fuel of misinformation. This research introduced an Xception-based approach to deepfake detection. The architecture utilizes an efficient feature extraction CNN model that benefits from the improvements such as depthwise separable convolutions and is optimized to exploit subtle artifacts in typical deepfake media, such as inconsistencies in facial landmarks, lighting irregularities, and unnatural textures. It was trained and tested on an all-inclusive dataset of more than 500,000 frames comprising authentic and manipulated video content. The data preprocessing steps included face extraction and alignment and diversity augmentation of the data to make generalization better. The model was evaluated using various metrics and scored quite impressively at 99.69% accuracy, precision at 99.58%, recall at 99.80%, and with an F1 score at 99.69%. These results demonstrate that the model is correctly balanced to avoid false positives at a reasonable tradeoff for identifying deepfake content well enough to be deployable in real-world scenarios where content verification will prove critical. The Xception-based model also shown strong computational efficiency with the average inference time being 0.08 seconds per frame. With an AUC-ROC score of 0.9999, this efficiency combined proves that the model can significantly distinguish the real from the manipulated frames with near-perfect accuracy. More so, its robustness was tested under various conditions, including different resolutions and compressions. It achieved an accuracy of 99.65% on low-resolution frames and 98.1% on high-resolution frames, with an accuracy of greater than 90% even for highly compressed videos that ascertained its adaptability to the diverse media qualities characteristic of content usually found on social media. In summary, a deepfake detection framework based on Xception offers a powerful, accurate, and efficient solution to deepfake media, supporting the integrity and authenticity of digital content on social media. As possible future directions of this work, it could be considered to further explore the adversarial training for more robustness against emerging deep fake generation techniques and expanding the dataset with occluded and diverse samples. This research is therefore a valuable contribution to the area of deep fake detection, offering scalability of the tool to reduce misinformation and protect public trust in digital media.
Enhanced energy demand management in electric distribution networks using lstm-xgboost model
Accurately forecasting electricity consumption is essential for improving the reliability, operational efficiency, and resilience of energy systems, particularly given the rising integration of renewables and the advancing complexity of smart grids. This research introduces an innovative multilayered model for short-term electric load forecasting that combines Long Short-Term Memory (LSTM) networks with Extreme Gradient Boosting (XGBoost) to enhance accuracy and dependability. The approach consists of three main steps: data cleaning and feature extraction, development of separate LSTM and XGBoost models, and combining them into a unified hybrid architecture. The Elia Grid dataset from Belgium was used in this study, containing high-resolution load data for 2022 captured at 15-minute intervals. The hybrid model leveraged the LSTM's strength in learning sequential dependencies, while XGBoost contributed by capturing non-linear residual patterns and extracting feature importance. The proposed model underwent extensive testing to evaluate its performance against independent LSTM and XGBoost models. The hybrid model achieved its best results with a Root Mean Square Error (RMSE) of 106.54 MW and Mean Absolute Percentage Error (MAPE) of 1.18% and Coefficient of Determination (R²) of 0.994. The sensitivity analysis showed that increasing the look-back window size improved model performance but the attention mechanisms did not enhance accuracy so they were removed from the final design. In addition to outperforming traditional models, the proposed framework demonstrated strong generalizability, scalability, and interpretability, making it suitable for real-time energy management systems. The model was benchmarked against recent classical and deep learning (DL) models and showed competitive or superior results across multiple datasets. This work contributes to the design of AI-based applications in smart grid management and enriches the literature on hybrid DL techniques for time series forecasting. Future research will explore expanding the model to multi-regional datasets, incorporating weather and socioeconomic variables, and deploying the system in real-world grid control environments.
Makine öğrenmeyi kullanarak COVID-19 aşı görüşlerinin duyarlılık analizi için sınıflandırma algoritmalarının karşılaştırmalı bir çalışması
The task of analyzing textual data and classifying them into positive, negative, or neutral emotions within the domain of natural language processing is a multifaceted undertaking. The primary objective of this study is to employ machine learning algorithms in order to classify opinions pertaining to the coronavirus disease and vaccines. In this study, four algorithms were employed, namely Random Forest (RF), Gradient Boosting Classifier (GBC), Logistic Regression (LR), and Decision Tree (DT). The RF and GBC algorithms demonstrated a commendable accuracy rate of 89%, while the LR and DT algorithms yielded a slightly lower accuracy rate of 87%. The findings derived from this research can provide valuable guidance to policymakers in effectively addressing potential barriers that may impede the successful execution of vaccination campaigns. The analysis of the Kaggle data, which encompasses a wide range of commentaries related to the pandemic and vaccines, underscores the urgent need for prompt measures to attain herd immunity against Covid-19. This imperative objective holds significant importance in effectively managing the transmission of the virus and mitigating its adverse consequences on the well-being of the general population. The task at hand necessitates the acknowledgment and resolution of public apprehensions, as well as the establishment of trust and assurance in the vaccination initiative. This study presents an analysis of the machine learning techniques employed and conducts a comparative evaluation of their significance. The forthcoming research endeavors to create an application that will be capable of categorizing sentiments and opinions pertaining to diseases and vaccines. It is imperative for governments and organizations to comprehend the obstacles linked to the worldwide COVID-19 vaccination endeavor in order to develop efficacious strategies. Nevertheless, it is crucial to acknowledge that the scope of the analysis was restricted to tweets written in the English language. This limitation may potentially undermine the credibility and generalizability of the findings pertaining to overall sentiment. Additional investigation could be conducted to examine more extensive Twitter datasets employing deep learning models in order to gain a deeper comprehension of the general public's attitudes towards COVID-19 vaccines.
Göğüs röntgeni görüntülerinde çocuk pnömonisinin otomatik olarak tespitinde derin öğrenme ve makine öğrenme yöntemlerinin etkinliğinin değerlendirilmesi
Accurate identification and classification of images of patients with pneumonia is vital for effective diagnosis and treatment. Advanced learning techniques such as CNN LSTM (Convolutional Neural Network Long Short-Term Memory) have proven 97% accuracy on this task. A labeled dataset of 5,856 images of pediatric pneumonia patients from Kaggle was used to evaluate these models. The dataset was divided into training (70%), testing (15%), and validation (15%) sets. It is worth noting that the ResNet 50 -CNN model and the CNN-ExtraTrees model achieved an accuracy rate of 98% and 94%, respectively. These models provide resources for clinicians and radiologists in their decision-making processes. Extracting features from images plays a role in pattern prediction using the optimizer as part of the modeling process. To improve performance, modifications were made to the RMSprop parameters. Model performance evaluation included metrics such as recall, precision, F1 score, and overall accuracy evaluation. These measures provide insight into the effectiveness of models while also serving as reference points for investigations. The exceptional accuracy rates achieved by these models confirm their ability to accurately classify patients with pneumonia, which has implications for improving patient outcomes. The evaluation criteria used in this research provide an examination of the advantages and limitations of models that highlights the importance of incorporating these metrics to ensure accurate performance in scenarios. Keywords: Medical Image, Pediatric Pneumonia, Resnet50-CNN, CNN-Extratrees, Convolutional Neural Network, Long Short-Term Memory
Alzheımer hastalığının gen seçimi için mikrodizi veri setleri ve mimari kullanılarak geliştirilmiş biyo-ilhamlı mantaray
Alzheimer's disease (AD), a prominent neurodegenerative condition, poses significant challenges in its early diagnosis and treatment due to its intricate pathophysiology characterized by the presence of amyloid plaques and neurofibrillary tangles. The recent developments in microarray technology, machine learning, artificial intelligence, and data mining analysis techniques hold considerable potential for comprehending the genetic foundations of AD. Nevertheless, the prediction of AD using machine learning faces several challenges surrounding the analysis of genes expression datasets. Summarized in the "curse of dimensionality" caused by the high dimensional disease's microarray datasets, the accurate prediction of disease can significantly suffer from overfitting, bias, and computational demands. To address this challenge, genes selection methods are applied to alleviate the effect of Alzheimer high dimensional genes expression datasets and elevate the overall machine performance.
Geliştirilmiş dikkat ve genişletilmişkonvolüsyon tabanlı topluluk model tabanlıağ saldırı tespit sistemi, geliştirilmiş cheetah optimizörü kullanarak düşmanca kaçınma saldırılarına karşı
A reliable defense system against network threats over a long time is called the Intrusion Detection System (IDS). Defenders are notified by the IDS when suspicious or malevolent activities are identified on the network. In the last ten years, machine learning has helped the IDS to become more accurate, more capable of analysis, and more adept at finding new or modified forms of known intrusions. Deep learning, an advanced version of the machine learning technique, is essential to the field of network security. Additionally, a deep learningbased Network Intrusion Detection System (NIDS) performs better than conventional IDS techniques. Recent studies, however, demonstrate that when faced with attackers in realtime, the deep learning-based IDS becomes somewhat inaccurate. There is no analysis done on how attack models will affect NIDS as well. In order to defend against adversarial evasion attacks, an enhanced deep learning-based NIDS model is designed here. The required data is first collected from commonly available websites. The best feature extraction is carried out on the collected data. Here, the Improved Cheetah Optimizer (ICO). Then, an Attention and Dilated-based Ensemble Network (ADEN) is implemented to detect the intrusions from the optimally extracted features. The Deep Temporal Convolutional Neural Network (DTCN), Long Short-term Memory (LSTM), and Gated Recurrent Unit (GRU) models are assembled together to deploy the suggested ADCEN. In the end, the ADEN detects the viii intrusions and generates the respective outputs using the fuzzy ranking approach. To demonstrate well the recommended deep learning-based NIDS defends against adversarial evasion assaults, experiments are conducted against conventional models.
Eliminate entanglement in quantum information processing utilizing hybrid quantum-classical neural networks
Quantum entanglement is a physical phenomenon that lies at the heart of the contrast between quantum physics and classical physics, which distinguishes quantum mechanics as an essential feature. It happens when a collection of particles interacts with one another, participates, or creates a collection of particles in close spatial proximity in a dependent way in which the quantum state of one particle in this collection cannot be characterized independently of the other particles' quantum states. The states of quantum mechanics have several types of order, including Symmetry Protected Topology (SPT), in which matter has symmetry and a finite energy gap at zero temperature. Feedback set methods are used to derive results (how to identify specific quantum data source attributes) in a more consistent manner. The concept of quantum entanglement refers to that SPT states are symmetric short-range entangled states., since only "trivial" topological orders characterize short-range entangled states. The researcher presents and analyzes: Convolutional neural networks inspired by a quantum circuit-based technique, which is highly effective when applied within a hybrid neural network environment allowing them to be efficiently trained and executed on realistic quantum devices, this research simplifies and implements a quantum convolutional neural network (QCNN) in a hybrid neural network environment on the TensorFlow platform, where it functions as a suggested quantum equivalent to a conventional convolutional neural network. This thesis demonstrates how to detect certain characteristics of a quantum data source, such as a device's intricate simulation or a quantum sensor. With or without an excitation, a cluster state will operate as the quantum data source, which the QCNN will learn to identify. In this study, the researcher classed the SPT phase as a dataset. The QCNN system attained a validation accuracy of 100% Instead of 89.58% in its purely quantitative analog, we use three layers of quantum convolution followed by a classical densely connected neural network to identify an exciting cluster state within 10 epochs of time. This architecture should be quite successful in mitigating entanglement.
Smart cities: A strategic approach to urban sustainability
Current trends in global urbanization and pressing issues around sustainability provide cities with considerable challenges. The concept of a "smart city" had been evolved like a method of dealing with cities in terms of the reality that they are becoming systematically more complicated due to interrelated structures and more reliance on the progress of communication and information technology to fulfil the demands of people. In this research, we examine the concept of "smart city" as a potential sort of city that might address the requirements and environmental sustainability issues of modern civilization. Smart cities are distinguished by advancements in their physical and human resources, as well as their traditional and modern telecommunications networks. These cities have a better lifestyle as well as a sustainable economic growth accomplished via democratic governance and the management of natural resources. Through in-depth interviews with smart city employees and sustainability specialists, the concept's benefits and drawbacks are uncovered and then organized according to (FSSD) the Framework for Strategic Sustainable Development. Then, a strategy referred to (SSD) as Strategic Sustainable Development, which is also a methodology, is used to maximize the value of the idea and reduce any recognized limits. This dissertation presents a planning guide guided by an SSD approach with the intention of supporting smart city projects in moving strategically toward their concept of a smart city while concurrently advancing society toward sustainability.
Tasarım ve performans analizi OPNET tabanlı WiMAX ağ IEEE 802.16 modelleyici
Because of the greatest coordination region, the unimportant expense of sending furthermore, and quick data rates, the WiMAX is a promising improvement for giving remote latest miles affiliation. Video Conferencing is one of the quickly making advances and ought to supplant the standard circuit exchanged video associations. Video Conferencing is in addition considered to be the first for these executioner apps of WiMAX that require a watchful plan for quality-of- service game plans. The important issue in this work for studying and examining this introduction of Video Conferencing in smaller WiMAX affiliations. The work had finished through the overview and assessment of several perspectives related to the WiMAX affiliation and Video Conferencing plans, for example, WiMAX association classes, adaptability, and the number of focuses. As far as possible coordinate throughput, and delay. The entertainment of various WiMAX network conditions utilizing OPNET Modeler. According to the results, the video conferencing has been calculated for traffic sent regarding the raising time was initialized at 109 seconds when the settled time reached at 13990000 bytes rate, while the traffic received raising at 110 seconds and the time of settled around 290000 bytes rate and the overall WiMAX delay is arranged for 0.31 raising delay for the first 100 seconds while the settled time that arrived to 1.25 for the overall 300 times as the end of the simulation.
A new biometric system using deep learning: Analysis of palmprint images
Rapid developments in technology can affect humanity both positively and negatively. Especially with the increase in crimes committed in cyber environments, important security studies on information security have been made in the world. For this reason, today, many companies and institutions take advantage of biometric recognition systems and use these devices in order to protect their security. In this study, new method based convolutional neural network pretrained model AlexNet and wavelet transform applied to recognize the humans. The proposed method extracted new and effective features from hand images and wired the extracted features to the ensemble learning classifier that classified the features into multiclass that are represented the 72 human. The obtained results 99.14 which is remarkable when compared with state of art studies.
Geometric analysis of pathological changes in the brain using mri images for tumor diagnosis
Any kind of camouflage treatment increases the chance of developing a brain tumor diagnosis analysis or another condition. Researchers have found common risk factors and symptoms of brain tumors after poring through massive amounts of data from individuals who have been diagnosed with the disease. New therapeutic approaches are being developed all hours of the day and night thanks to advances in scientific technology. The doctor may discover and diagnose a brain tumor by looking at this photograph. Consciousness loss is a major risk factor for the development of brain tumor diagnosis (Segmentation, measurements, detection, and classification), and this trend is only expected to grow. This study utilizes an algorithm to forecast disease risk specifics based on the detected tumor region in brain MRI scans and measurements, and the work focuses on determining the tumor's range and form. A tumor is an abnormal, uncontrolled development of tissue. There are several tumor varieties, each with its own unique characteristics and therapeutic approaches. Due to the nature of the tumor and the small size of the cerebral cavity, brain tumors are very dangerous and potentially fatal (space formed inside the skull). According to studies of causes of death, brain tumor deaths are disproportionately high in wealthy nations.It accurately forecasts disease details from the tumor location and establishes the stage and size of the tumor. Segmentation techniques for brain tumors are the combination of fuzzy c-means and thresholding used here. This technique enables tumor tissue segmentation with precision and consistency on par with that hand segmentation. To insure the medical report, we have done a color map of the brain tumor and non-infected area. When testing the proposed framework on MRI and CT images, the tumor identification rate was about 96%, with a DSC of 97 % and an IoU of 96 for automated and manual brain tumor segmentation.
Optimization of the number and placement of routers in wireless mesh networks
Wireless mesh networks are a new type of wireless network that has been growing in popularity. These networks consist of routers and clients. the routers are called mesh routers and the clients are called mesh clients. Wireless mesh networks have a number of advantages over traditional wireless networks, such as more reliable coverage and faster speeds. One of the disadvantages to these networks is that they require more router placement than traditional wireless networks. This is because each router needs to be within range of all other routers in the network. There are many different types of algorithms that can be used to determine the best placement for these routers, with some algorithms being better than others depending on the environment or situation. One algorithm is called a Genetic Algorithm which uses a process called genetic programming to find an optimal solution for router placement. (GA) genetic algorithm is used to find the best placement of the router so that it can provide the most coverage possible for a specific area. It uses a probabilistic approach and avoids placing too many nodes in one spot, which can lead to connectivity issues. The suggested method was implemented utilizing the C++ programming environment and the NS2 software suite. Using a benchmark of produced instances, the experimental outcomes have been analyzed. Variable sets of produced instances ranging in size from small to big have been explored. Consequently, several properties of WMNs, including the topological placement of mesh clients, have been recorded. In terms of network connection, coverage, and a number of routers, simulation results also indicate that superior performance is attainable with the suggested strategy.
Predicting the likelihood of a patient's survival after an accident using artificial intelligence and statistical methods
The goal of this research is to construct a model that, with the help of regression methods and deep learning, will be able to compute the probability of survival for patients who are undergoing medical care in hospitals at the present time and who are exhibiting symptoms that are typically seen in intensive care units. These patients are considered to be in a high-risk category (ICUs). Survival models are a type of statistical tool that are used relatively frequently in the field of clinical research. Clinical researchers are interested in determining how long a patient is expected to live. Researchers in clinical settings may make use of survival models to the purpose of these models is to forecast the risk of a variety of clinical outcomes as well as to identify relevant risk variables. A good illustration of this would be the overall survival rate of patients who suffer from a variety of diseases. According to the findings, our model is capable of performing better than the CPH prediction model when it comes to projecting the prognosis of patients who have CCU. This is the case when it comes to the prognosis of patients who have CCU.
Blood glucose microwave sensor based optically control detection
The development of a microwave resonator design for noninvasive blood glucose biological sensing is the focus of this thesis. The proposed sensor is built upon a printed transmission line with a characteristic impedance of 50. The diffusion-lines have been overtaken by a sweep penny elegance chaotic' meander line. The construction is divided into two symmetric pieces, which are joined to one another by a resistor that changes in resistance in response to light (LDR). In order to increase the field fringing through the sample being examined, an interdigital capacitor is inserted in parallel with the meander line configuration. It is discovered that the proposed sensor offers numerous frequency resonances. The authors used frequency resonance at 3.46GHz to ensure that the detection would work effectively with fluctuating blood glucose levels. In order to comprehend the effects of altering the blood glucose level on the transmission loss spectra of the proposed sensor, which are measured in terms of S11 magnitude, the proposed sensor is developed and experimentally tested with 100 patients at the international/Applied &Theoretical/Research-Center (IATRC). The performance of the suggested sensor is evaluated and enhanced using circuit model theory. Lastly, the finite integral approach algorithm in the commercial software program Computer/Simulation/Technology of Micro/Wave/Studio is used to numerically validate the design performance (CSTMWS). In the end, encouraging experimental and theoretical discoveries are made.
Neural network and model predictive control for synchronous generator system
Today's recovery in the whole world seems to come with the high momentum towards electrical energy as a direction of industrialization. This project, which is electricity generation, is the right choice of generators used as backup and industrial facilities, which are a more important part of facility expenditures. In this study, new method based neural network technique LSTM and particle swarm optimization algorithms presented to deal with this problem. The aim of this study predictive control for synchronous generator system. In the first stage, several techniques such as support vector machine, random forest, and decision tree applied to this problem. Then, the obtained results are not suitable which lead us to developed new technique-based LSTM. The results of classical LSTM obtained and then the LSTM based PSO applied to deal with this problem which is new study. The obtained results are compared and show that the proposed LSTM based pso is best than other studies
COVID-19 diagnosis with artificial intelligence algorithms
When the covid-19 virus initially spread globally in 2019, As a result, numerous people passed away. Due to this sickness, medical staff members are busier than ever. For reducing this load, computer-aided diagnostic tools and machine learning will assist in some ways. Any scientific investigation into this disease can help it be eliminated rapidly. In our study, we utilized the Covid-19 dataset which is publicly accessible (14486 cases) from the Kaggle website. Training and test data categories were created from the dataset, using 80% of the data for training and 20% for testing. This research has two main stages: First, we extract Visual Geometry Group (VGG16) features from the dataset, then based on these extracted features, five machine learning algorithms were employed for classification (k-nearest neighbor (KNN), random forest (RF), extreme gradient boosting (XGBoost), decision trees (DT), and support vector machine (SVM)). Second, we extract gray-level co-occurrence matrix (GLCM) features from the dataset. Then, we applied the same five machine learning algorithms (DT, RF, KNN, XGBoost, and SVM) for classification based on these extracted features. Performance metrics were examined utilizing a confusion matrix, accuracy, recall, and precision together with the F1 score.Regarding classification accuracy, the Support vector machine algorithm and VGG16 were the classification techniques that excel at 98.22% accuracy and achieve the impressively rapid and best result.
Design of a microstrip diplexer for recent wireless communication systems
In recent years, microstrip diplexer filter has gained much recognition, especially in wireless communication systems like mobile phones, radio transmittals, broadband wireless, as well as satellite-based communication systems, due to their ability to combine two separate filters in one device using a communication channel. In this work, a meander line resonator, a steps impedance resonator, and a uniform impedance resonator are used to build a microstrip diplexer using FR4 epoxy laminate material. Mathematical formulas are used in the design process to zero in on the optimal qualities that will allow for fine-tuning the resonant frequency. The meandering structure is proposed to miniaturize the suggested diplexer compared to its original size. High isolation between the output ports is accomplished, as well as the structure maintains a basic topology and small size. From the proposed design, excellent performance characteristics including high isolation, low insertion loss, and good frequency selectivity are all achieved. The proposed design has been created and stimulated using Advanced Wave Research (AWR) has been at 1.94GHz and 2.515GHz for GSM-Ⅱ, WIFI/WLAN. An FR4 epoxy laminate is used to fabricate the suggested diplexer, and the dimension of 32×26mm^2, with an insertion loss of 2.9 dB and a return loss of 31.69 dB; the simulated isolation between both two filters is greater than 25.2 dB. Finally, the design sample has been implemented by fabrication to verify the simulated results. The fabricated diplexer has been measured using an analyzer VNA (vector network analyzer).
Development and implementation of yolov8-based model for human and animal detection during forest fires
For the biosphere to be protected, forests are a necessity everywhere in the world. Forest fires are one of the major risks to life in many parts of the world; they put the environment, including humans, plants, animals, and even land in danger. The North African & Mediterranean regions, Amazon & Australia last year suffered greatly from forest fires. To save lives and property, forest fires must be discovered sooner rather than later. This study aims to detect fire, smoke, humans, and animals in outdoor images using the YOLOv8 deep learning algorithm. A custom dataset of outdoor images was created by obtaining images from various search engines and manually annotating them. The YOLOv8 model was trained on this dataset and achieved an overall mAP of 0.274, with varying performance for different object classes. The model had difficulty in detecting small instances of fire and smoke and struggled to differentiate between animals and humans in certain cases. The study also identified the importance of image quality in computer vision and highlighted the impact of poor image quality on model performance. Overall, the study presents a comprehensive evaluation of YOLOv8's performance in detecting outdoor objects and identifies areas for improvement.
Examining the potential of deep learning in the early diagnosis of Alzheimer's disease using brain MRI images
Millions of people around the world are affected by Alzheimer Diseases, which is a major public health issue worldwide. The key to effective treatment and management techniques is early diagnosis of the illness. For the purpose of detecting Alzheimer Diseases using MRI data, we looked into and evaluated three distinct DL models in this study. The first model used is CNN with two convolutional and two fully connected layers served as the initial model. The second model was an improved version of the first, with a leaky ReLU activation function, more fully connected layers, and a larger kernel size. The third model was a transfer learning model with two dense layers that was built on top of the VGG16 architecture. An extensive set of MRI scans from Alzheimer's patients and healthy controls was used to train the models. The first and second CNN models achieved an accuracy of 96%, while the transfer learning model achieved an accuracy of 81%, according to the accuracy, precision, recall, and F1-score measurements. In conclusion, MRI data-based Alzheimer's diagnosis may benefit from DL models. However, further progress is required to improve these models' performance and accessibility for clinical use.
Development of AODV security against black hole attack in mobile AD HOC network
Security has emerged as one of the most critical concerns to be addressed in MANET (Mobile Ad-hoc Network). Because of the high speed of MANET and the huge amount of nodes, security moved to be a major complicated and too hard challenge in the network, causing it to be unprotected from various kinds of security attacks. In this blackhole attacks were used to be able to evaluate their effects on the MANET. The blackhole attacks are investigated in both single and multiple attack scenarios. They are simulated using NS2 simulator version 2.35 for network simulation. The AODV (standing for Ad-hoc On-Demand Distance Vector) protocol was used to send data messages from one node to another one. Four sets of evaluation used cases were performed in two networks type (static, and dynamic) with (10,20,40, and 60) nodes and speeds ranging from (20-100) MS and implementing appropriate solutions. The R-AODV (Reverse-AODV) and ACO (AntNet Colony Optimization) algorithms are used to eliminate the effect of single blackhole attacks and to improve the MANET under the effect of multiple blackhole attacks, respectively. The full amount of the dropped packets, the ratio of the delivery packet, and the throughputs are calculated and compared to show the change in the efficiency of the AODV protocol improves after implementing solutions to eliminate the impact of blackhole attacks on the network (MANET). The findings show that the attacks of blackhole have a greater effect on AODV. The efficiency of the network (MANET) is highly affected by the number of nodes and nodes' moving speed, in our performed simulation, network behaviour shows better performance in less number of nodes and high speed. After reviewing the previous studies that related to Ad-hoc networks generally and attacks that target those networks especially I choose the blackhole attack as subject for my study as the blackhole attacks has a higher effect on AODV protocol compared to other attacks like flooding and rushing attacks, from my end I suggest two possible solutions that can help in eliminate blackhole attacks that I will explain it in details and practices in this thesis. R-AODV has a greater impact on reducing attack effect than AntNet solutions for a single blackhole attack.
Algılanmazlığı artırmak için fıbonaccı ayrışması ile kontrast değişimine dayanan geliştirilmiş bir steganografi sistemi
There are currently many obstacles in the way of the design and development of a reliable image steganography system. These include low capacity, weak robustness, and invisibility. Overcoming these restrictions requires enhancing the steganography system's capacity and security while keeping the signal-to-noise ratio high (PSNR). In light of these considerations, the purpose of this research is to create a technique to successfully embed secret data into a cover image, thereby realizing a strong steganography scheme. The planning and execution of the suggested method occurred in multiple stages. To boost the scheme's text security and payload capacity, a novel encryption approach dubbed shuffle the segments of secret message was integrated with an improved Huffman compression algorithm. To further strengthen the approach, the bit depth of each pixel was doubled from 8 to 12 using a Fibonacci-based picture transformation decomposition method. Third, the schemes stealthiest was bolstered by the use of an enhanced embedding technique by combining a random block/pixel selection with the implicit secret key generation. Experimental evaluations of the suggested scheme's performance are conducted to determine its stealthiest, security, robustness, and capacity. Against the proposed scheme, resistance is analysed for their resistance to non-structural, 2, and statistical steganography detection attacks. The acquired PSNR values indicated that the proposed technique was successful in achieving the higher imperceptibility and security than the reported findings while preserving the larger capacity. In a nutshell, the problems were viii fixed since the proposed steganography system was superior to existing data hiding schemes on the market.
Renewable energy utilization in demand-side energy management system based on linear programming optimization algorithm
Demand-side management (DSM) is an effective strategy in smart grid technology to manage energy demand and promoting the efficient use of energy by regulating appliance scheduling and regularly monitoring energy consumption. By integrating DSM into residential energy management, consumers can be empowered to manage their energy consumption and reduce peak demand, which in turn reduces the need for expensive new production and transmission infrastructure. This approach also helps to reduce carbon emissions and enhance grid reliability. To further enhance DSM, optimization techniques can be used to manage large-scale appliances with varying power ratings. By modelling the energy system as a set of linear equations, these algorithms can determine the optimal use of available resources and minimize waste. Linear programming optimization algorithms are a key component of these systems in addition, optimization system based on forecasted cost, power, and energy. By analysing this data, the system can predict when energy prices will be high or low and adjust energy consumption accordingly. This achieved by suggested more than one scenario to test system reliability and used MATLAB/Simulink to simulate the proposed system taking into account various factors such as weather conditions, energy demand, and pricing trends. Compared to other methods using various optimization algorithms, the suggested approach used for lowering electrical energy costs in a microgrid system while maintaining their regular load and operating hours. By minimizing the energy cost of the community, the algorithm can determine the optimal energy consumption pattern that meets the community's energy needs while minimizing costs. Efficient DSM optimization algorithms can significantly enhance the smart grid's efficiency while reducing electricity costs for consumers. Ultimately, the integration of DSM optimization algorithms can help to achieve a more sustainable and cost-effective energy future.