Theses supervised by Prof. Dr. Ahmet Zengin
12 theses · Sakarya University
A new forecasting system for electrical loads in the middle euphrates region using neural networks
This study develops neural network models for electrical load forecasting in Babylon Governorate, Iraq, focusing on Feedforward Neural Networks (FNN) and Long Short-Term Memory (LSTM) networks. Load forecasting is essential for managing power supplies and systems because it can guide the optimal allocation of resources, planning the right time for maintenance and reducing energy costs among others. Load forecasting is of great importance due to the complexity of power systems and the variability in demand. Statistical methods such as ARIMA have been used in traditional forecasting techniques. Although they work well in some circumstances, they do not incorporate all non-linear structures between variables. To address this, neural networks, especially those based on deep learning, have proven useful tools by enhancing forecasting accuracy. This thesis aims to use "FNN and LSTM", a form of artificial neural networks to forecast electrical load demand. The data analyzed in this research includes load data for 33/11 kV power distribution stations in Babil Governorate, Iraq, from January 2021 to April 2024. Temperature information was also collected from reliable sources such as Accuweather, Wunderground, and Weather. Load data and weather conditions were integrated into the model. The main activity is to forecast the electrical load for 2024 and 2025 with special attention to the effects of retaining fluctuations in demand during distinct months. Some of the data used in this research include monthly electrical load demand records as well as temperature records during the same period. The load data consists of energy consumption of several feeding stations located throughout the governorate, and temperature data is also used to understand the difference in energy use with the help of weather conditions. The first process was data cleaning and relevant data were formatted to improve data quality. MinMaxScaler and standardization methods were used where the data were normalized to scale for input into the neural network model. Normalization was applied, specifically, to make the ranges of all features equal because it is mandatory for the convergence of neurons during training. Model Design and Training, two different artificial neural network models were applied: the feedforward neural network (FNN) and the long short-term memory (LSTM) network. The aim is to analyze data and predict outcomes based on the nature of the available datasets. The selection of the optimal model depends on several factors, including data structure, model accuracy, and training efficiency. The FNN and LSTM models were compared, and loss and validation plots were constructed by tracing their respective loss curves during training and validation. The model was constructed, and model summaries were created. The FNN consists of dense layers with dropout regularization to prevent overfitting, resulting in a relatively lightweight architecture (approximately 11,521 parameters). However, the LSTM includes specialized layers for sequential data (such as LSTM and embedding), making it more complex (approximately 79,375 parameters). This complexity enhances its suitability for temporal or textual data but requires greater computational resources than the FNN. To validate and avoid overtraining the model, the Early Stopping technique was used to monitor the performance and stop training when the performance improvement stopped. The data obtained was divided into training and test datasets, where 80% of the data was trained and 20% was tested. Special methods, including the Dropout layer, were introduced to enhance the generality so that the model does not depend too much on the patterns in the training database. Moreover, hyperparameter tuning was also performed to fine-tune the "model performance," including learning rate, batch size, and number of epochs. The prediction accuracy of the neural network model constructed in the present work was measured by Mean Square Error in the test data (MSE) Mean Absolute Error (MAE) and Mean Percent Absolute Error (MAPE). These metrics provide a good indicator to evaluate the model's ability to predict unseen data and identify areas that deserve further improvement. The first set of results indicated that the model captured the general behavior of the electrical load demand pattern but with varying degrees of accuracy in different months. For example, there were higher demands during the summer, especially in July and August, as people used more air conditioning in various residential facilities. Similarly, demand in the winter months of January and February was higher due to heating systems. However, as temperatures dropped in April, May, October and November, the modeled demand was observed to adjust or decline. The model was particularly sensitive to these seasonal differences, hence the need to incorporate correct weather data into subsequent versions. Our results show a large approximation to the true values, differences in some months due to climate change, and large differences between summer and winter temperatures; temperatures range from a maximum of around 50 degrees in summer to a minimum of 5 degrees in winter. This is a large difference. This resulted in the test data evaluation being an approximate MSE of 95,000, MAPE of 18% in the test data evaluation. This indicates that the model can highly accurately predict and is useful in capturing broader trends. There are certain months that it is expected to produce lower forecast errors, and more factors should be considered when developing the model. More complex neural network architectures should be included in future studies to obtain better results due to some advantages of processing long series. Including more variables across multiple sectors, such as economic indicators, energy prices and consumption patterns, would make the demand model more accurate. The techniques used to produce ensemble estimates are generally superior to individual models because they can reduce variance and bias and thus produce more reliable estimates. Although the model has been shown to perform well, work and development in this area always need to be developed to keep pace with the increase in electrical loads and human growth; the model can be improved by incorporating neural network structures such as recurrent neural network / long short-term memory (RNN/LSTM) that can handle time series such as electrical load demand. These structures could cope with long-term dependencies in the data, which is necessary for forecasting over a long period. Another recommendation is to try adding more variables, including standard of living, economic factors, energy costs and energy consumption rates by different residential, industrial and commercial sectors. The first set of results indicated that the model captured the general behavior of the electrical load demand pattern but with varying degrees of accuracy in different months. For example, there were higher demands during the summer, especially in July and August, as people used more air conditioning in various residential facilities. Similarly, demand in the winter months of January and February was higher due to heating systems. However, as temperatures dropped in April, May, October and November, the modeled demand was observed to adjust or decline. The model was particularly sensitive to these seasonal differences, hence the need to incorporate correct weather data into subsequent versions.
A surveillance and security alert system based on real-time motion detection to protect secret and vital places
In the modern world, surveillance systems are crucial to the security sector. Video surveillance systems have traditionally employed moving object detection extensively. In computer vision, surveillance systems have grown in popularity recently. In this study, we have developed a system to protect vital and secret places from thieves. It should also facilitate the work of areas that require monitoring systems because security concerns have exploded in the current environment. The system relies on human motion detection. The main objective of the research is to create a system that can be part of reducing and eliminating theft and evil and provide an easy system with a user-friendly interface at a low cost. Cameras have been used to detect human movement and monitor areas, allowing observers to monitor areas directly through the system. Whenever the system detects human movement through a camera, it directly notifies the observers of the presence of human movement in areas where there should be no human movement through sound alerts, LED alerts, and emails to the observer. It starts videotaping the area, as well as recording and storing the time, place, and photo of the movement to make it easier for observers to follow up. The system consists of two main parts: software and Arduino hardware. In the software part, the types of algorithms used were the two-frame difference and simple background modeling for movement detection with the assistance of AForge.NET and C# and Tiny YOLOv2 for human detection. In the Arduino hardware section, Arduino is used to create an LED alert. Evaluation and testing of the system show that it can be used indoors and outdoors and detect human movement in all four directions, such as the back, forward, right, and left. In low light, the system has the ability to determine human movement. The experiments tell us that the system's accuracy in determining human motion was 95%, and depending on this result shows that the system can preserve vital and secret locations and correctly surveil zones. Future work includes adding alerts, such as mobile calls to observers through the system, which will help control incidents as soon as possible. Adding a mobile application to the system, where observers will be able to monitor the areas via their mobile phones, may allow them to monitor the areas from anywhere. Also, it will be possible to detect fire and smoke in areas where surveillance cameras are installed. And it is possible to use other Yolo models for human detection and other motion detection methods in the surveillance and security alert systems.
Medical image compression based on vector quantization and discrete wavelet transform
Due to the prevalence of chronic diseases, diagnostic imaging procedures are increasing annually. However, storage is challenging because storage capacity is limited to accommodate these ever-increasing medical images. As a result, it has become important to develop new ways to compress these images due to their large sizes. Therefore, there is a need for a compression method that preserves medical diagnoses and has a high compression ratio. Image compression aims to significantly reduce the size of different types of images while preventing distortion and harmful distortion during reconstruction. The medical image compression system is an important topic in image processing systems to reliably compress and decompress various types of medical images since image compression has become the latest emerging trend all over the world. This can be accurately implemented for both lossy and transparency techniques, such as discrete waveform transform (DWT), discrete cosine transform (DCT), fractal-based and wavelet-based notation vector quantization (VQ), predictive coding, and so on. Although these techniques are used to design the compression system, they face some challenges, such as reducing the original size, level of computational complexity, and minimum square of compression errors. This thesis addresses these critical challenges to improve medical image compression performance. There are two main ways to compress images, each of which is determined by whether or not the original image can be reconstructed from the compressed image. Lossless compression is where no information is lost in compression or decompression operations, and this is used with medical images to keep all image information. Lossy compression is where the entire image cannot be decompressed because it uses a lossy compression method where some information has to be deleted. To compress digital images without losing quality, they are widely used along with JPEG. The purpose of this thesis is to introduce a simple and effective image compression technique that increases the efficiency of the compression system and improves the performance of compression methods based on discrete waveform transform (DWT) while maintaining the mean structural similarity (MSSIM) and Peak signal-to-noise ratio (PSNR) at an acceptable level. Medical images represented by MRI, CT, and ultrasound are the data sources for this thesis. We used MATLAB (R2022b) to formulate the necessary coding. To accomplish this work and to take advantage of the capabilities of the program in the field of image processing, this thesis proposes a combination of Discrete Wavelet Transform (DWT) and Vector Quantization (VQ) to process complex medical images while preserving the diagnostic content. Medical images inevitably suffer from salt and pepper noise; in this method, filtering removes them during pre-processing. We use [3×3] Gaussian algorithm to determine the value of each pixel in the output image based on the value of its associated pixel and adjacent pixels in the input image. Then wavelet analysis (DWT) can divide image information into approximate and detailed partial signals. A close-up sub-signal shows the general direction of the pixel value, and three detailed sub-signals show vertical, horizontal, and diagonal details or changes in the image. If this detail is small, it can be set to zero without significantly changing the image. There are two types of filters, high pass filter and low pass filter. Thus, the signal is effectively divided into a detailed high-frequency part and a rough low-frequency part. A Discrete Wavelet Transform (DWT) is applied to preserve the edges of the images (horizontal, vertical and diagonal) as the image is divided into four bands, and the sub-band conversion is repeated with the highest level of specificity. Vector Quantization (VQ) technique is used for data compression, correction, pattern recognition, density and clustering estimation. Missing data from some ranges is retrieved by finding the closest set, including the available data ranges. It will get the same value as the midpoint VQ applied to all the parameters generated by the previous step except for the sub-range. This contains the most important details for image recovery since the basis of VQ work is to represent pixel arrays by an index in the codebook. Many improvements are made to reduce the computation time for image compression. The experiments performed at this point are different codebook sizes within different-sized windows for each image. Performance efficiency was evaluated with different sizes of codebooks and windows and for increasing the pressure by arithmetic coding. Then we use the threshold principle between the two methods that generate wave coefficients and exclude the value less than the specified threshold. The best combination of hybrid technology was found at two levels of DWT before quantification. This thesis tested the images by applying three separate analytical levels of the wavelet transform. The resulting wave coefficients were then encoded using image coding techniques based on wavelet transform, which is arithmetic coding. With this proposed hybrid technology, we achieve a good performance, and by adding an operation optimization algorithm, we achieve even better performance. The proposed compression scheme achieved better results than the referenced studies and increased the compression ratio. The test showed that decomposing the wave into three levels achieved the best compression ratio and the best encoding technique used with the previous wave. Several factors affect the results of the previous tests, the most important of which are the type of encoding technique for wavelet coefficients and the optimization algorithm used. Tests showed that decomposing the wavelets into three levels achieved the best compression ratio and the best encoding technique used with the previous wave. To compress images, we use the arithmetic coding (AC) algorithm, which uses fractions between 0 and 1 to generate a unique string of characters. A data code encoding and decoding algorithm repeatedly processes and encodes (decodes) a single data code, making it symbolically iterative. When the algorithm iterates, it divides the period between 0 and 1 into smaller and smaller parts, keeping one of these pieces as a new period. The technology runs on layered time slots, each of which contains a sequence of code that is evaluated for size. Code stream size comparisons are used to reconstruct the data stream, indicating how the encoder divides and stores overlapping subintervals, unlike well-known compression methods such as prefix Huffman symbols. Then we turn to artificial intelligence, using the particle swarm optimization (PSO) algorithm. The particle swarm development method is almost certain to find a globally optimal solution. The global search ability is also great, and the computational performance is much faster than traditional methods. The test was done using the Compression Ratio (CR) parameter, the Mean Square Error (MSR) parameter, and the Peak Signal to Noise Ratio (PSNR) parameter. Experiments were performed with different codebook sizes and window sizes for each image. A test was performed to evaluate the efficiency of the measuring tool with different sizes of codebooks and windows. Experiments at both levels confirm the larger size of the codebooks. The best are chosen to maintain image quality and thus a suitable choice of the hybrid algorithm because image quality is critical in medical imaging. At the end of the work, we compared it with a reference study where the fine image details were used in terms of extension, dimensions, and pixel depth, which gave better results than the referenced studies and increased the compression ratio. It was found that applying the proposed scheme to compress images using optimization algorithms helps to obtain the best results Providing several proposals to take into account the time and improve performance.
Yazılım test sürecinde hata yönetimi ve yeni bir hata yönetimi uygulamasının geliştirilmesi
Yazılım Yaşam Döngüsü, bir yazılım işinin veya projesinin fikir aşamasından itibaren analiz, geliştirme, test ve canlı ortama çıkış aşamalarını kapsayan bir süreçtir. Yazılım Testi ise bu süreçte yapılan yazılımın doğrulanma işlemlerini içeren önemli kısmını oluşturmaktadır. Yazılım Testi genel olarak Hata Yönetimi araçları ile yönetilmektedir. Bu süreç boyunca araçlar kullanıldığında birçok paydaşın olduğu bu uzun sürecin izlenebilirliği sağlanır ve güçlü bir raporlama avantajı sağlar. Birçok marka veya kurum bu süreçler için araç kullanımını tercih etmektedir. Araçların çoğu tek başına yetersiz kalmaktadır. Bu durum için çeşitli çözümler bulunmaktadır; Kullanılan araca uyumlu başka araçlarla eksiklikler tamamlanabilir. Açık kaynak kodlu araçlar üzerinde geliştirme yapılabilir veya kendi süreçlerine uygun araçlar geliştirilebilir. Bu tezde halihazırda kullanılan bir hata yönetimi aracı tecrübesinden yola çıkılmıştır. Araç ile ilgili çalışanlardan gelen olumsuz geri bildirimler ve öneriler değerlendirilmiş ve süreci daha pratik hale getiren yeni bir uygulama geliştirilmiştir. Uygulamada testin ana girdisi olan gereksinimler test edilebilir gereksinimler olarak alınmıştır. Gereksinim ekranında hata girişi sağlanmıştır. Gereksinimler, test senaryolarına dönüştürülerek aynı ekran içinde test durumları gösterilmiştir. Süreçler, mevcuttan ve çeşitli araçlardan çok daha kısa hale getirilmiştir. 39 adımda yapılan bir hata raporlama işlemi 8 adıma indirilmiştir. Buna ek olarak geliştirilen uygulamanın karmaşık olmaması, ayrı bir giriş ekranı olmaması, gereksinimlerle senaryoların aynı ekran üzerinde görülebilmesi, hata şablonu oluşturulabilmesi hata girişi maliyetini yaklaşık %80 azaltmıştır. Ayrıca çalışanın motivasyonunu artırması uygulamayı daha verimli hale getirecektir. Geliştirilen hata yönetiminde mevcut sistemde olmayan aşağıdaki özellikler bulunmaktadır. Kullanıcı ve yetki tanımlama, lisans ve bakım ücreti, araç eğitimi bu uygulamada olmayacaktır. Mevcut sistemdeki kullanım zorluğu en aza indirgenmiştir. Senaryolar otomatik olarak gereksinimlerden üretilmektedir. Hata kayıtlarında şablon oluşturulmaktadır. Hata kaydı ekranına işletim sistemi versiyonu, internet tarayıcı çeşidi gibi bir çok gerekli detay bilgiler eklenmiştir. İstatistiksel bilgiler tutulmasını kolaylaştıran bu ek geliştirmeler genel iyileştirmeyi sağlayacaktır.
A new shortest path algorithm for management urban traffic
Nowadays, traffic jam is an interminable obstacle for the transportation growth in urban cities around the world. One of the major causes of traffic jam is raising of vehicles number due increment population, mainly in areas with bottlenecks and it also causes for society and economic losses, increase in greenhouse emissions and health damages. Furthermore, it targets communities in the most critical element of life which is the human element. Therefore, it has become essential to work to find solutions, projects and put them into practice to address the causes and mitigate the negative effects. This thesis focuses on improved Dijkstra algorithm based on traffic congestion level. Improved Dijkstra algorithm can provide (a) real data collection from map via OpenStreetMap (b) Adding four features to SUMO simulator software (time period, rush-hour, number of vehicles, and routing algorithm) (c) it has the ability to know congestion level for roads (d) rerouting vehicles to avoided traffic congestion. Based on the simulation results and analysis presented in the thesis, it was found that the proposed improved Dijkstra algorithm increased the performance of the road traffic flow by reducing the number of related vehicles in traffic congestion and average delay time for experiments scenarios. The improved Dijkstra algorithm has decreased related vehicles and delay time by approximately 23%. On the other hand, the improved Dijkstra reduce average fuel consumption and amount of CO2 emission by approximately 15% and 14% respectively. The future works is to examine the performance of improved Dijkstra algorithm on the real road traffic environment and to employ (VANTE) technology in vehicle route guidance systems, which will play a vital role in alleviate traffic congestion by exchanging traffic information between vehicle to vehicle and the vehicle to infrastructure. The contributions and findings of this thesis may support urban planners, mobile operators, transport planners, civil engineers, and traffic congestion management researchers and in addition reduce environmental pollution.
Development of data encryption algorithm for database security by using ASCII code
Currently, many people apply technologies, which are gradually moving towards new progress and development. Moreover, many of us utilize database systems to assist our work management; the database contains our work or personal information, which increases the risk of losing our data due to disruptive electronic attacks. As a result, protecting databases from electronic attacks and data seizures is crucial, and one method of identifying data is through several algorithms so that no benefit is taken from our data during electronic attacks. In this paper, we explain a particular formula we created for the Encryption of data. The data is encrypted using Ascii Code. Also, we used three keys in the main formula. Because of that formula, each data we save in the database will be encrypted. The data can be Text or number. And through using another coordinator with the three previous keys, we can render the data to our original style. The formula focuses on data size and recording speed in such a way that we get the same data size as when the data is encrypted at a reasonable speed.
Design and implementation of a new blockchain algorithm to increase reliability, security and integrity
The utilization of blockchain is increasing day by day because of its extra ordinary features including distributed and decentralized storage services. Blockchain can provide flexibility, tamper resistance, immutability, fairness, transparency, and robustness. Moreover, the addition of smart contract provides increases the programmability and management facilities. Although blockchain was first introduced to support cryptocurrency, special facilities make it popular for different fields like e-commerce, global payments, P2P landing, remittance, healthcare, record keeping, voting, logistics, etc. Smart devices including internet of things (IoT), internet of vehicles (IoV), internet of healthcare (IoH), etc. also start utilizing blockchain for different purposes. However, too much flow of data and duplication increase the scalability problem and there is no efficient solution available to minimize this problem. Thus, in this thesis, we proposed a novel multi-level blockchain structure to minimize the scalability problem. The system is divided into two parts which are global and local blockchain. Local nodes are the member of a local blockchain where all the local service center is a member of the global blockchain. Global blockchain stores information of all the local blockchains'members. Local service centers will provide their support to only the local members and global blockchain will be used to handle the migration process. Because of the proposed structure, local blockchains will not be overloaded and thus able to perform more efficiently and quickly. To implement the proposed structure, we used Vehicular ad hoc networks (VANET). Smart vehicles while moving around can form a temporary communication with the nearby vehicles to form a VANET to create social networking between them. Blockchain is used by researchers to ensure the security and authenticity of the vehicles, store and analyze traffic events, and also manage and distribute the transmitted messages. However, almost all of them suffered from scalability problem. To minimize this problem, in this thesis, we use blockchain to manage the authenticity and message transmission of both cluster-based and co-operative VANET. Four different systems have proposed in this thesis and implemented in the ethereum blockchain platform and programmed by using smart contracts. Simulation results and performance analysis shows that the proposed methods provide security, integrity, authenticity, tamper free, robustness as well as outperforms previously available systems.
Sürücülerde uykululuk hali tespit sistemi tasarımı ve uygulaması
Uykululuk halinin önceden tespit edilmesi, uykululuğa bağlı trafik kazalarını önlemek, fiziksel ve ekonomik kayıpların önüne geçmek açısından önemlidir. Bir kamera yardımı ile kişilerin görüntüleri üzerinde görüntü işleme teknikleri uygulanarak uykululuk halini önceden kestirmek mümkündür. Bu tezde, literatürde kullaılan uykululuk ölçme yöntemleri, veri kümeleri ve görüntü işleme tekniği incelenmiş ve uyarlamalı bir eşik değeri hesaplama yöntemi önerilmiştir. Veri kümelerinden yararlanılarak uykululuk halinin tespit edilmesinde büyük bilgiye sahip olan göz bölgesinden gelen öznitelikler ile göz kırpma ve uzun süreli göz kapama tespiti yapmak için sabit ve uyarlamalı eşik değerleri ayrı ayrı değerlendirilmiştir. Böylece, kısa süreli göz kırpma ile uzun süreli göz kapamanın daha iyi ayırt edilmesi hedeflenmiştir. Çalışmada önerilen uyarlamalı eşik değerinin sabit bir eşik değerinden çok daha başarılı göz kırpma tespiti sonuçları verdiği, iki farklı veri kümesi üzerinde yapılan deneyler ile doğrulanmıştır. Uykululuk tespiti için literatürde CLOSDUR ve PERCLOS yöntemleri kullanılmıştır. Uyarlamalı eşik değeri kullanılarak elde edilen göz açıklığı bilgisi ile gözün bir dakika için kapalı olduğu sürenin yüzdesel ifadesi (PERCLOS) belirlenmiş ve uzun süreli göz kapama bilgisi (CLOSDUR) ile birleştirilerek uykululuk tespiti çalışamaları yapılmıştır. Elde edilen sonuçlar literatürde bulunan veri kümelerinin gerçek referans değerleriyle büyük oranda uyuşmaktadır.
Siber güvenlik analizi için yeni bir siber saldırı simülatörü geliştirilmesi
Kurumların ve bireylerin iş süreçlerini bilişim teknolojileri ile yürütme zorunluluğu beraberinde risk ve tehditleri de getirmiştir. Siber saldırılar, kurtarılması zor sonuçlara yol açabilir. Bu saldırılara karşı pek çok saldırı tespit ve güvenlik sistemi geliştirilmiş olsa da, bilgi sistemlerine yönelik saldırılar ve güvenlik ihlalleri hızla artmaktadır. Bu çalışmada siber güvenliğin sağlanması noktasında en önemli konulardan biri olan güvenlik zafiyetlerinin anlaşılması ve siber saldırıların tespit edilmesi amaçlanmaktadır. Siber saldırı yöntemlerini test etmek için fiziksel ağları kullanmak çok maliyetli ve zaman alıcı bir süreçtir. Bu tez çalışmasında, siber saldırı senaryolarını simüle etmek, test etmek ve sonuçları değerlendirmek için DEVS modelleme yaklaşımı kullanılarak bir siber saldırı simülasyon modeli geliştirilmiştir. Saldırı modellerinin üzerinde çalıştıralacak model ağın topolojisi, BRITE topoloji üreteci tarafından oluşturulmuştur. Saldırı modllerinin özelliklerine göre saldırı senaryolarının parametre ve yapılandırma ayarlarının yapıldığı görsel arayüzler kullanılarak saldırı simülasyonu gerçekleştirilmiştir. Saldırı adımlarının etkilerinin ve sonuçlarının gözlemlenmesi ve değerlendirilmesi için simülasyon ve izleme çerçevesi kullanılmıştır. Bir sanal ağda bir saldırı senaryosunu simüle eden ve uygun saldırı tespit sistemi uyarıları üreterek bu uyarıları değerlendiren bir uygulama geliştirilmiştir. Bu çalışmada, saldırı önleme amacıyla ağ trafiğinin incelendiği ve şüpheli etkinlik belirlendiğinde "refleks" tipi eylemlerin gerçekleştirildiği saldırı tespit sistemlerinin metodolojisine benzer bir yöntem geliştirilmiştir. Sistemin test edilmesi için Kanada Siber Güvenlik Enstitüsü tarafından paylaşılan CSE-CIC-IDS2018 veri seti kullanılmıştır. Geliştirme ortamı olarak DEVS-Suite simülasyon paketi kullanılmıştır. Farklı siber saldırı simülasyon uygulamaları ile karşılaştırmalar yapılmış ve farklılıkları ortaya konmuştur. Bu araç belirli saldırı türleri için uyarı verileri elde etmek için kullanılsa da, sonraki çalışmalarda daha farklı saldırı türleri için de uyarı verileri oluşturmak üzere genişletilebilir bir altyapı sağlamaktadır. Bu çalışmada, büyük ölçekli kurumsal ağların kolaylıkla tasarlanabileceği ve geçerli düzeyde performans, ölçeklenebilirlik ve doğruluk ile siber güvenlik testlerinin kısa sürede yapılabileceği görülmüştür.
A new deep learning based object detection system for increasing salesman performance
Food distribution companies must be flexible and responsive to provide the best services to their customers. Operational costs and expenses are the biggest concern for these companies, as they rely on many employees to serve customers and achieve the company's goals. Increasing tasks creates continuous and growing pressure on the sales team and the rest of its supporting departments, which affects the time and effort of the salesperson in achieving daily tasks and thus achieving the expected monthly goal. Deep learning technologies provide solutions to help reduce the effort expended by the sales team and, at the same time, contribute to reducing operating costs. In this thesis, an integrated system is proposed to use deep learning techniques to reduce the representative's effort and compensate the human support staff with deep learning algorithms to provide the necessary support to the representative. YOLO algorithms are used to discover objects in real-time and Faster-RCNN algorithm is used to solve planogram issues on shelf. The results extracted from images are used from these algorithms and converted into data stored in the MySQL database. Google API techniques are used to transfer the results from the cloud to the company's servers and thus analyze the results received using the Microsoft Power BI tool. The results show that proposed system provides a complete solution to reduce the effort of the sales team and significantly reduce costs. Also, this system can be used to study the competitor productions visibility on shelves.
An IoT based smart irrigation recommendation system
The largest proportion of fresh water around the world is consumed in agriculture. As a result, there has been a continuous interest in proposing systems and solutions that achieve rationalization of water resources in agriculture without negatively affecting productivity. On the contrary, the solutions provided must take into account raising and improving productivity by using the least amount of water resources. The Internet of Things is one of the prominent recent technologies that contribute to providing many solutions in most fields, including agriculture in general and irrigation specifically. This study proposes a framework for smart irrigation based on the Internet of Things, by building a prototype consisting of a control unit, water pumps, and several sensors. These devices are used to measure the soil's need for irrigation and to determine the appropriate amount of irrigation according to sensor data. As these values are sent through cloud computing to a mobile application that is installed on the user's mobile to allow the possibility of monitoring irrigation and controlling it from several aspects. This includes manual irrigation mode, determining the value at which irrigation is done automatically in automatic irrigation mode, and the appropriate amount of irrigation according to the relationship between different sensors. This thesis made several contributions, including the adoption of the dynamic feature, whether in making the irrigation decision according to the type of plant or in determining the appropriate amount of irrigation according to the relationship between the sensor readings of each plant. In addition, records of irrigation operations are stored in the automated mode for the purposes of feedback and future improvement on the proposed system. In future work, other sensors such as rain sensor and water level sensor will be added to increase the automation of the proposed model, in addition to using artificial intelligence and deep learning techniques to make irrigation decisions and determine the appropriate amount based on the data that is already stored in the current proposed system.
Tracking a person through the camera on the internet and determining the location
I declare that the thesis work titled "TRACKING A PERSON THROUGH THE CAMERA ON THE INTERNET AND DETERMINING THE LOCATION", which I have prepared in accordance with Sakarya University Graduate School of Natural and Applied Sciences regulations and Higher Education Institutions Scientific Research and Publication Ethics Directive, belongs to me, is an original work, I have acted in accordance with the regulations and directives mentioned above at all stages of my study, I did not get the innovations and results contained in the thesis from anywhere else, I duly cited the references for the works I used in my thesis, I did not submit this thesis to another scientific committee for academic purposes and to obtain a title, in accordance with the articles 9/2 and 22/2 of the Sakarya University Graduate EducationandTraining Regulationpublished in theOfficialGazettedated20.04.2016, a report was received in accordance with the criteria determined by the graduate school using the plagiarism software program to which Sakarya University is a subscriber, I have received an ethics committee approval document, I accept all kinds of legal responsibility that may arise in case of a situation contrary to this statement.