Search

In this study, it is aimed to produce meaningful words in accordance with character-based Turkish grammar rules by using encoder-decoder and attention architecture, which are deep learning algorithms. The results of the developed model are compared with the results of LSTM and GRU models, which are other deep learning algorithms. It is seen that the language models created with LSTM and GRU models give similar results at 100 and 200 epoch values and at different threshold values of the temperature sampling method. Among these models, the GRU model gives the highest success value with 88.40% at 200 epochs and 0.5 temperature threshold value. The encoder-decoder and attention language model developed for this study gives the highest success value of 91.90% at 100 and 200 epoch values and at different threshold values of the temperature sampling method at 200 epoch and 0.5 temperature threshold value. At the end of the experiments, the encoder-decoder and attention architecture model showed an average of 2.83% more success than the LSTM model and an average of 0.19% more success than the GRU model. [ABSTRACT FROM AUTHOR]

Firms in the manufacturing sector need to constantly monitor their performance in order to maintain their development under competitive conditions in the market. In this study, eleven KPIs are determined to measure the production performance by taking into account the factory assets. The proposed system is designed in which the relevant KPIs are obtained via the instantaneous data received from the CNC machines in a production system. The main objective of this study is to measure and estimate production performance. In this way, it is aimed to provide a proactive approach for the assets whose performance is monitored by the decision-makers. LSTM and LightGBM models which are deep learning techniques are proposed for the estimation of performance indicators. The approximately three-month time series OEE (Overall Equipment Effectiveness) values of the sample CNC machine are used for estimation. The estimation performance of methods is compared over performance metrics (MSE, MAE, etc.). The results indicated that LightGBM outperforms LSTM for all performance metrics. [ABSTRACT FROM AUTHOR]

Thanks to the precision agriculture technologies developed in recent years, many processes such as irrigation, fertilization, spraying, weeding and harvesting of agricultural products can be done by autonomous systems. Especially in some plant species such as sunflower, when to apply these processes is largely decided according to the developmental stage of the plant. In this study, deep learning methods were used to classify the developmental stages of sunflower plants. Since the images taken with the drone are high resolution, each of them is divided into 6 equal parts, and then 8 classes are determined and the images belonging to each class are extracted. A data set consisting of 12800 images in total, 1600 in each class, was created. Six different deep learning models, namely AlexNet, InceptionV3, ResNet101, DenseNet121, MobileNet and Xception, were tested with Sgd, Adam and Rmsprop optimization methods and their performances were compared. In order to evaluate the success of the models correctly, the trained models were also tested on a second data set created with images taken from a different terrain and high success rates were obtained. In addition, a 7-class test set was created for images that could not be clearly determined at which stage the plant was in, and the success rates of the models were tested. It was observed that the success rate was very low for the images in the 7-8 intermediate class, and the filters used in the image processing techniques that would increase the success rate for this class were applied to the images, and the models were retrained and the results were evaluated. [ABSTRACT FROM AUTHOR]

After lung cancer, prostate cancer is the most common type of cancer in men. Nowadays, advanced prostate imaging is conducted by multiparametric prostate magnetic resonance imaging (MRI) performed by radiologists. Although many pathologies of the prostate can be visualized, the main purpose is to determine the probability of prostate cancer and to decide whether a biopsy is needed. This process includes the analysis of different series of Magnetic Resonance (MR) images, including T2-weighted images (T2W), diffuse-weighted images (DWI), and dynamic contrast enhanced images (DCE). However, firstly it is necessary to differentiate the prostate zones. Then, lesion scanning is performed in the relevant prostate zones. Finally, there is a need to perform prostate lesion scoring according to PI-RADS v2. Therefore, diagnosis of prostate cancer is a complex and long process. For this reason, decision support systems are needed for the diagnosis of prostate cancer. In this context, the main purpose of the study is to provide automatic segmentation of prostate regions. In the segmentation task, Mask R-CNN algorithm was used with T2W MR images of 15 patients. With the use of the ResNet-50 backbone model, the Mask R-CNN algorithm achieved the segmentation performance with a mAP50 value of 96.040. Finally, the trained model was integrated into the PACS system. Thanks to the integration, a ready-to-use in hospitals artificial intelligence-supported decision support system was developed. Thus, time was saved while reducing the workload on health employees. [ABSTRACT FROM AUTHOR]

Diagnosis of anomalies in the gastrointestinal tract is a current research area. Wireless capsule endoscopy (WCE) for the evaluation of this region is a preferred alternative technology to avoid the risks of traditional endoscopy and to provide a painless process. But this technology, which has many advantages, offers low frame density. This situation affects the quality of the data and causes to a decrease in the diagnostic accuracy rate. In this study, WCE endoscopy images obtained from KID Atlas Dataset 2 were used and a three-stage artificial intelligence-supported diagnostic process was developed for the detection of the inflammatory anomaly, vascular anomaly, polypoid anomaly and normal image categories in the gastrointestinal tract. For the first stage, critical points on the images were clarified using 5 different approaches. These improved images were classified with a region proposal-based object recognition algorithm and performance comparison was made according to the approaches used. In the second stage, the data augmentation technique was applied to the improved images that showed maximum performance in the first stage. Thus, a dataset with a balanced and sufficient number of images was created. In the third stage, this current dataset was classified with five different object recognition algorithms. However, the individual success of each algorithm is different. For this reason, the ensemble learning approach was used to obtain stable outputs for each category and to create a balanced detection process among the categories. Finally, a balanced and stable estimation function was provided between categories with this hybrid structure. [ABSTRACT FROM AUTHOR]

Hydrothermal alteration is an important geological feature used in the exploration stages of precious minerals. This research focuses on two distinct deep-learning network structures created to identify hydrothermal alteration types in microscope images. A dataset of 2500 images, 70% of this data set was used to train, 20% to test, and 10% to measure the validity of the network. Convolutional Neural Network (CNN) and Xception models were trained using Adam, RMSprop and SGD optimization functions and the results are discussed. The Adam and SGD optimization functions for the CNN model performed the most successful classification with 96% accuracy. In the case of the Xception model, the highest accuracy value was 98% for the networks using the Adam and RMSprop optimization functions. Although the Xception model had higher accuracy values, it was observed that the CNN model completed the process significantly faster considering the training time of the network. [ABSTRACT FROM AUTHOR]

It has become imperative to monitor the data-driven industrial systems of today's technology before potential failures occur. Predictive maintenance predicts these failures before they occur and takes the necessary action to prevent malfunctions from occurring. In this study a comparative predictive maintenance application which is based on machine and deep learning is realized. Logistic Regression, Naive Bayes Classifier, Decision Tree, Support Vector Machine, Random Forest, and K-Nearest Neighborhood are used as the classical machine learning methods while Long Short-Term Memory and Gated Recurrent Unit are used as the deep learning architectures. The performances of the methods are examined on the Predictive Maintenance dataset from UCI Machine Learning Repository for fault type detection and the results are presented comparatively in terms of metrics in detail. In the experimental studies, fault type detection is handled separately in the form of multiple and binary classification problems. In the solution of the multiclassification problem, the highest accuracy among the machine learning methods is obtained by Random Forest method with 98.26%, while the accuracy value obtained with both deep learning architectures is 97.51%. In the solution of the binary classification problem, after the data balancing, the highest accuracy among the machine learning methods is obtained by Random Forest method with 95.03%, while the highest accuracy among the deep learning architectures is obtained by Gated Recurrent Unit architecture as 93.03%. [ABSTRACT FROM AUTHOR]

Nowadays, deep learning methods are widely used in many fields such as robotics, sound processing, medicine and imagery. In this study, studies on image classification and analysis in the literature were examined in detail. In addition, performance analyzes of deep learning methods were carried out using CNN on 4 different data sets named MNIST, Fashion MNIST, CIFAR-10 and CIFAR-100. The structures of the created deep learning models, the parameter values used for training, the layers used, the complexity matrices obtained for the validation data, accuracy and loss graphs are shown in detail. Our study was carried out with a different network structure consisting of 3 convolution layers, 3 batch normalization layers, 2 maxpooling layers, 1 flatten, 2 droupout, 2 dense layers. In addition, the performance of image classification on different data has been increased with our proposed model. At the end of the test, the accuracy results of our study were compared according to various evaluation metrics. For all the data sets used, the best and worst images were determined. With the proposed CNN model, high accuracy rates were observed for MNIST and Fashion MNIST datasets, with these values being 99.22% and 99.21%, respectively. [ABSTRACT FROM AUTHOR]

Today, the use of face recognition systems and attacks against these systems have increased. Especially with the increasing use of social media, the increase in the sharing of facial images and videos allows attackers to deceive facial recognition systems more easily. For this reason, face spoofing detection has become a very important field of study. Face spoofing attacks are carried out in various types. Generally, performance evaluations are made on scenarios in which all attack types are evaluated together. Therefore, in this study, face spoofing detection performances of deep learning methods in Printed Photo, Digital Photo and Replay Video attack types in the Replay-Attack dataset were evaluated. For this purpose, the face spoofing detection performances of VGG16, DenseNet121 and MobileNet deep network architectures were examined. Then, the change in face spoofing detection performances because of classification of deep features produced by each network with support vector machines (SVM) was examined. Finally, the deep features produced by VGG16, DenseNet121 and MobileNet networks were combined (feature level fusion) and real/fake classification was performed for all attack types with SVM. According to the results, the classification of deep features or their combinations with SVM increases the performance of face spoofing detection according to the attack type. [ABSTRACT FROM AUTHOR]

Nowadays, the Internet is a structure that people can access easily and at the same time produce content easily and without control. In parallel with this situation, the ability of extract information from the raw data that makes up big data has become more complex. The fact that the headlines of the contents contain uncontrolled and misleading elements makes it difficult to reach the right information. The headlines of the contents are important for people to reach the information they want in their limited time. In this study, it is aimed to produce headlines suitable for the content instead of headlines that may be misleading for news. For this purpose, an application that produces headlines for Turkish news with deep learning method has been developed. SuDer news corpus is used as dataset. For the training of the model, it is aimed to obtain more humanoid results in the production of news headlines by using the Transformer architecture, which is frequently preferred in natural language studies today and the abstract summarization method. In this study, in order to compare the performance of the Transformer model, models are prepared and trained with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures. At the end of 25 epochs of training with LSTM, GRU and Transformer architectures on the corpus, the values of loss are 1.03, 0.55 and 2.49 respectively. In the experiments performed on the validation data, measurements are made with ROUGE-1, ROUGE-2 and ROUGE-L metrics. As a result of the measurements, it is observed that the Transformer architecture is partially good, based on the metric values produced. In addition, when the headlines produced with these architectures are examined, it is observed that the headline obtained with the Transformer architecture produce headlines that are partially more suitable for the news content compared to other architectures. [ABSTRACT FROM AUTHOR]

In this study, the Deep Learning technique was combined with the Uncertainty Quantification scale to determine whether patients have cancer or not, and if cancer is present, to classify it as benign or malignant using a dataset of 663,549 lung tomography data. The reliability of this model was also investigated. 80% of the mentioned images were used for training, 10% for testing, and 10% for validation. Two methods were proposed for this purpose. In the first method, measurements were made without using Uncertainty Quantification, and the accuracy rate for the ResNet50 model was found to be 0.965, while the accuracy rate for the AlexNet model was calculated as 0.837. Additionally, the accuracy rate for the Inception v3 model was found to be 0.849, and for the VGG16 model, it was 0.826. Firstly, the accuracy rate of the 4-Layer Convolutional Neural Network model was calculated as 0.971. In the second method, measurements obtained using Uncertainty Quantification (UQ) resulted in an accuracy rate of 0.89 for the Bayesian Neural Networks ResNet50 model. Furthermore, the accuracy rate for the Bayesian Neural Networks AlexNet model was 0.86, for the Inception v3 model it was 0.91, and for the VGG16 model, it was 0.77. The study has innovative features such as the development of a new model called the 4-Layer Convolutional Neural Network and the use of uncertainty quantification methods. These results demonstrate that the performance of specific neural network models can be enhanced by considering uncertainty. [ABSTRACT FROM AUTHOR]

Firms that operate in the production and service sectors take various decisions for the future under uncertainty in order to combat the increasing competitive conditions. One of these critical decisions is sales forecasting. With the spread of digital technologies, the use of deep learning approaches in sales forecasting is increasing. Although deep learning gives successful results, it needs long training time with large amounts of data. As a solution to this situation, transfer learning (TL), which provides information transfer between problems, is used. Transfer learning provides the training of the model with source data and its transfer to the target data. In this study, a deep transfer learning approach is proposed to transfer the information obtained from the sales forecasting models of different products to future forecasting models. Sales data are considered as univariate time series. The edit distance with real penalty (ERP) between the target and source data is used as a measure of transferability in the selection of source data. A long short-term memory (LSTM) network has been trained, which enables modeling of temporal dependencies with the selected source data. ERP-LSTM-TL forecasting model is created for the target data by transferring parameters from the pre-trained LSTM network. In the experimental studies on sales data from various industries, compared to LSTM trained with target data, ERP-LSTM-TL provided average 5.77% and 7.42% improvements in terms of root mean squared error and mean absolute error measures, respectively, also it shortened training time by average 47.90%. ERP-LSTM-TL gave statistically better results compared to classical forecasting and machine learning methods. [ABSTRACT FROM AUTHOR]

Nowadays, with the increased use of computers, a surge in data production has emerged, making data access more convenient. In this context, a substantial amount of textual data is generated on e-commerce sites, social media, and various electronic platforms. Analyzing and extracting meaningful insights from this amassed data proves valuable for numerous institutions, organizations, and individuals. Sentiment analysis is a commonly employed technique to derive sentiments from textual data, and contemporary sentiment analysis models often leverage the high performance offered by deep learning approaches. Prior to model training, several pre-processing steps are typically applied to the text data. In this study, three distinct deep learning approaches were proposed for sentiment analysis. These models were analyzed on two different datasets: winvoker and Beyazperde. Hyper-parameters and depth of models were optimized using the Bayesian optimization method to enhance the accuracy of model. Additionally, the impact of various pre-processing techniques on model performance were assessed. When non-preprocessed data is utilized, the models trained on the winvoker dataset achieve an accuracy of 94.16%, while those trained on the Beyazperde dataset reach 86.64%. With the application of pre-processing, these accuracies improve to 94.64% for the winvoker dataset and 89.08% for the Beyazperde dataset. Based on these findings, it was concluded that the effect of pre-processing decreased and the accuracy was higher for the winvoker data set with a higher number of samples. [ABSTRACT FROM AUTHOR]

In this study, a study was conducted with the HAM10000 dataset containing skin lesions to see the performance of architectures such as EfficienNetB3, which uses deep learning techniques to detect skin cancer. In this study, applications were made in deep learning to see the effect of the learning rate or learning rate on architectures using convolutional neural networks used in image classification. These applications were made to see the effect of the learning rate, the size and diversity of the data set, and the number of images used in training, as well as the increased number of images, on both the success in classification and the time taken for training. EfficientNetB3 was used as the architecture and HAM1000 dataset, which is open access on the Kaggle platform, was used as the dataset. At the end of the study, without using as many augmented images as possible and with 600 images of each disease class, a learning rate of 0.002 and epoch 15 instead of 10, the highest accuracy performance of 0.8234 was achieved in the fifth application. [ABSTRACT FROM AUTHOR]

As seen in research conducted in the world and in Turkey in recent years, stomach cancer ranks first compared to other types of cancer. It is known as a type of cancer that is very difficult to detect. Gastric dysplasia is when the cells lining the inner surface of the stomach stop growing and developing normally. These cells begin to grow and multiply uncontrollably, then turn into stomach cancer. The main goal is to minimize this disease before it turns into cancer. Gastric dysplasia is divided into two types: low-grade and high-grade dysplasia. High-grade dysplasia is a precursor to cancer. Therefore, early diagnosis is extremely important for the diagnosis and treatment of gastric dysplasia. This study aims to detect dysplasia at an early stage by using deep learning and convolutional neural networks techniques for the detection of gastric dysplasia. In the study, feature extraction was performed on histopathological images using multiple Convolutional Neural Networks (ESA) models. The data, divided into two label classes: patient and healthy, were compared with the help of deep learning architectures and classifiers. [ABSTRACT FROM AUTHOR]

Dental health has a significant impact on overall health and quality of life. Segmentation of impacted teeth is a critical step for early diagnosis and treatment in dentistry. In this study, we investigate the use of deep learning techniques to accurately identify impacted teeth in panoramic dental images. In this context, a Pyramid Scene Segmentation Network (PSPNet) based on ResNet backbone network is developed for embedded tooth segmentation. In the proposed architecture, ResNet18, ResNet34, ResNet50, ResNet101 and ResNet152 versions of the pre-trained ResNet backbone network are adapted. Considering the findings obtained in this study, the ResNet18 model achieved the highest success in segmentation and recognition processes in dental images (92.09% F1 Score, 93.88% Precision, 90.39% Recall, 85.34% IoU Score and 96.89% Dice Coefficient). This research reveals that the rate of successful detection of impacted teeth in adult patients is high as a result of studies on panoramic dental images. These findings emphasize that AI can be an effective tool for dentists and increase confidence in the development of AI in the healthcare sector. [ABSTRACT FROM AUTHOR]

Objective: The aim of this study is to provide a different perspective on the analysis of RNA sequencing data by employing popular machine learning and deep learning methods, rather than classical statistical approaches. Additionally, it aims to provide insights into machine learning and deep learning concepts. Material and Methods: Utilizing machine learning and deep learning techniques, two distinct raw datasets pertaining to asthma and kidney transplantation (GSE85567 and GSE129166) were retrieved from the National Center for Biotechnology Information database and subsequently subjected to requisite quality control and alignment procedures. Random forest (RF), support vector machines (SVM), and deep neural networks (DNN) models were implemented to achieve patient-control differentiation. To prevent overfitting, all data sets were divided into 67.5% training, 10% testing, and 22.5% validation data, and 10-fold cross-validation was employed during the training stages of the models. Python programming language was used for both machine learning and deep learning, and Unix operating (AT&T Bell Laboratories, USA) system was utilized for data processing. Results: In the GSE129166 data set, the RF model obtained an accuracy rate of 0.89 in the validation set. The precision and recall of this model were determined as 0.88 and 0.92, respectively. The SVM model measured an accuracy rate of 0.88 in the validation set, and 0.87 in the test set. For the GSE85567 data set, the accuracy rate of the RF model in the validation set was measured as 0.73. For SVM, the accuracy rate in the validation set was measured as 0.70, while for DNN, it was measured as 0.75. Conclusion: The study conducted on the GSE85567 data set demonstrates that RF and SVM models exhibit high accuracy and performance. The DNN model, on the other hand, has a more balanced precision and recall rate, and is observed to be a significant alternative. Additionally, it is observed that the DNN model shows effective performance on the GSE129166 data set. Particularly, a high accuracy rate and a balanced precision-recall balance were observed in the validation set. It is concluded that all three models are suitable for patient-control classification in RNA-seq data. [ABSTRACT FROM AUTHOR]

Recognition of handwritten digits has recently gained importance and attracted the attention of many scientists, as it is used in many machine learning, deep learning and computer vision applications. Hyperparameter optimization involves determining a set of values aimed at increasing accuracy in both classification and prediction. It is also aimed to optimize the performance in feature selection by regulating the parameters selected by the algorithms more accurately. In this study, a convolutional neural network was used to recognize handwritten digits using the MNIST dataset. There are many open source hyperparameter libraries that deep learning developers can use to determine hyperparameters. In the developed model, hyperparameter optimization techniques were applied using Optuna, HyperOpt and Scikitoptimize libraries and the results were evaluated. Optimization times for hyperparameter libraries and the change in the success rate in recognizing handwritten digits were analyzed. The model trained with randomly given parameters achieved 78.45%, 97.13%, 75.62%, 76.95%, 97.46% and 97.27% accuracy, while the model trained with optimized hyperparameters achieved 99.26% accuracy. [ABSTRACT FROM AUTHOR]

With the rapid advancement of technology, the use of artificial intelligence (AI) is increasing in various sectors such as education, health, security and defense. A critical application of AI is highway management, especially with the rise of autonomous vehicles. The focus of this study is to address the problem of deformations in highway marking lines that pose challenges for autonomous vehicles and affect traffic safety. The research involves using unmanned aerial vehicle (UAV) to create an original image dataset of highway lines. This data set will be processed with image enhancement techniques and deep learning models. The first stage involves cleaning the images from foreign matter. Deep learning models will then identify potential line deformations. These models will be developed and trained for optimum accuracy using various performance metrics. In the study, the mobilenet v3 model, trained with the images in the data set, reached an accuracy rate of 89.58%, the resnet50 v2 model 77.78% and the Convolutional Neural Network model 92.55%. The ultimate goal is to implement a real-time system to accurately detect and report differences in highway lines by combining the UAV with the computer system on the ground. This will ensure timely notification to authorities and help prevent traffic safety problems related to line deformations. This approach demonstrates the practical applications of artificial intelligence in improving road safety and autonomous vehicle navigation. [ABSTRACT FROM AUTHOR]

Cracks in buildings are a source of concern as they may indicate structural problems. Cracks in buildings are one of the major problems with concrete structures, as they affect the appearance of the building, compromise the integrity of the masonry, jeopardize building safety, and reduce the durability of the structure. Cracks in buildings can be a cause for concern and may indicate a potential structural problem that could jeopardize the safety and stability of the building. Understanding the root causes of these cracks is crucial to determining appropriate preventive measures and repair methods. In this study, crack and slope control in buildings was performed with an image processing-based measurement system developed using CNN deep learning algorithms. A dataset of 294 photographs was used for the study. The data were subjected to image processing, preprocessing, and detection with CNN. The accuracy rate is 0.9716 for the VGG-16 architecture and 0.9513 for the Inception-V3 architecture. It was concluded that there is no significant difference between the accuracy rates of the architectures. All vertical and horizontal cracks with a thickness of 2 mm and a continuity of 4 cm were detected. Thus, crack control in buildings can be done quickly and reliably, and building safety will be ensured. [ABSTRACT FROM AUTHOR]

Rice, obtained through the processing of paddy, is one of the most widely consumed food products globally. However, diseases affecting rice plants, particularly those occurring on the rice leaves, pose significant challenges for farmers. The identification of rice plant diseases demands specialized knowledge, making it a complex issue to tackle. Often, due to insufficient understanding, farmers misdiagnose diseases and apply incorrect treatments. Rapid and accurate disease diagnosis plays a pivotal role in enhancing healthy and productive crop cultivation. To address this problem, a deep learning-based model was developed to detect rice plant diseases. The model was trained on a dataset containing four different rice plant diseases and achieved a successful outcome with a loss value of 0.0014. Additionally, four different deep learning algorithms were used to create models through transfer learning with pre-trained ImageNet models, and a comparison of their performance was presented. The most successful model was obtained using the VGG16 transfer learning architecture. Experimental results in this study demonstrate that the proposed transfer learning method can effectively recognize rice leaf diseases, providing a reliable approach for identifying leaf diseases in various plants. [ABSTRACT FROM AUTHOR]

The physical properties of fruit and vegetable products play an important role in their quality classification. With the existing manual methods, disease, pesticide and quality status of agricultural products are checked during the control stages. Manual sorting and classification of products requires expertise and the process is time consuming and labor intensive. Today, with the developing technology, the processing and marketing of products can be realized in optimum time and efficiency with the software techniques used in the fields of agriculture and food. In this study, date palm fruit, which has an important share in the fruit and vegetable market, is considered. The use of image processing and artificial intelligence techniques in classifying the quality of date fruits makes the sales process more consistent and efficient. Within the scope of the study, four different artificial intelligence techniques were used with a uniquely prepared dataset. The dataset consists of three different classes: good, bad and medium quality dates. The dataset specially prepared for the study was trained with MobileNetV2, ResNetV2-50, DenseNet- 201 and Inception-v3 deep learning models. Among the four models used in the study, MobileNetV2 model was determined as the most successful model with 95% accuracy rate. [ABSTRACT FROM AUTHOR]

With the rapid advancement of drone technologies, the use of drones has become of significant importance, particularly raising serious concerns in the areas of security and privacy. Deep learning and transfer learning artificial intelligence techniques hold promise for drone detection. However, to successfully apply these techniques, the need to develop new and efficient solutions for accurately detecting complex aerial conditions, variable speeds, and highly maneuverable drones is inevitable. In this study, the performance of training models using the EfficientNet model for drone detection was compared, and the challenges encountered were discussed, offering a perspective on potential future successes. According to the obtained results, when more layers are frozen in the transfer learning method, the required GPU memory for training decreases, leading to reduced GPU usage. This indicates that models trained with larger image sizes can be trained more quickly. Deep learning methods require more data and GPU resources, which extend the training time. In the conducted experiments, the best success rate achieved by a model trained with the deep learning method was 97.3%, while the model trained using the transfer learning method achieved the highest success rate of 99.7%. This demonstrates that transfer learning achieves higher accuracy with less data. However, the success rate obtained through the deep learning method is also considered quite satisfactory. [ABSTRACT FROM AUTHOR]

Classification studies on image data sets with convolutional neural networks algorithm, which is one of the deep learning approaches, are widely performed successfully in many fields such as medicine and agriculture. However, in cases where the sample numbers of the classes in the image datasets are imbalanced, the classification performance of this algorithm is negatively affected. In general, unlike the majority class, the minority class(es) are not well learned by the convolutional neural network algorithm. In such cases, applying oversampling methods provides successful results. With oversampling methods, the number of minority class(s) samples is increased, making it close to or equal to the sample number of the majority class. In this study, translation, rotation, random erasing, noise injection, mixing of images, kernel filters, generative adversarial networks, flipping, feature space transformation, cropping and color space transformation oversampling methods frequently used in the literature were applied on the Ham10000 dataset. According to the results obtained as a result of the study, oversampling methods have been compared in terms of classification performance. From three different convolutional neural network models, according to the classification results obtained with ResNet50, DenseNet201, MobileNet, in terms of accuracy, the noise injection method in the ResNet50 model was 0.967, the color space transformation method in the DenseNet201 model was 0.965, and in the MobileNet model, the mixing of images method had a classification performance value of 0.974 which was better than the other oversampling methods. [ABSTRACT FROM AUTHOR]

During apple (Malus communis L.) harvesting, physical damage that reduces the quality of the product is inevitable. Early detection and separation of damaged fruits is important in terms of increasing their commercial value. Undetected defective products reduce the production volume as well as food loss, since they affect the quality of intact products. The aim of this study is to detect defects in apples using deep learning techniques on images taken from the "Granny Smith" apple cultivar. A technique that does not require special conditions and that will make classification and defect detection cost-effectively has been researched. In the study, the test accuracy of the InceptionV3 model was 100% after 100 epochs, and the test accuracy of the AlexNet model was 98.33%. A method has been developed that can prevent economic losses that may occur after harvesting by detecting and separating the damages that occur on the fruit during harvesting with deep learning techniques. [ABSTRACT FROM AUTHOR]

Stroke is a cerebrovascular health problem that occurs when blood flow to the brain is slowed or interrupted for various reasons. Since the brain tissue cannot receive enough nutrients and oxygen, brain cells begin to die within minutes and temporary or permanent damage occurs in the functions in the area where the stroke occurred. Brain stroke is a very serious medical condition and requires urgent intervention. Early detection of stroke and segmentation of the stroke site are of great importance in terms of preventing permanent damage. In this study, brain stroke segmentation was performed using U-Net deep learning model with Res2Net backbone. A total of 2223 CT images including 1093 hemorrhagic and 1130 ischemic stroke types were used as dataset. The images were preprocessed with the windowing method and then trained and tested with the proposed model. While the IoU rate was 0.82 in the images used without windowing adjustment, this rate increased to 0.87 after preprocessing, when the data duplication method was added, the average IoU reached 0.92. The test results obtained showed that the use of appropriate window settings in the images increased the segmentation performance. [ABSTRACT FROM AUTHOR]

Detection and tracking of enemy targets is vital to military operations. In this study, the use of deep learning techniques to detect camouflaged tanks is examined. The main purpose of this study is to detect camouflaged tanks hidden in forest areas. In this way, the detection and tracking of enemy targets can be done more effectively and the security of the soldiers can be ensured. The YOLO architecture combines object detection and classification into a single network, resulting in faster and more accurate results. YOLOv5, YOLOv6, YOLOv7 and YOLOv8 architectural designs are compared. The YOLOv6 architectural design outperforms other designs. This design obtained 0,983 mAP50 value, 0,966 precision, 0,935 recall and 0,950 F1 score as a result of performance analysis using 2.234 tank images and 774 car images. The results of this study show that YOLO architectural designs show high performance in object detection and classification. [ABSTRACT FROM AUTHOR]

There are about 466 million hearing and speech disabilities in the world, and there are about 1.5 million hearing and speech disabilities in Turkey. Sign language (SL) is the language that is used by hearing and speech-impaired individuals to communicate among themselves and with other individuals. This language realized with hand, face, and body gestures is a dynamic language that can vary from country to country and from region to region in the countries where it is spoken. Learning sign language at a younger age is important, especially in terms of both cognitive and intellectual development and the academic development of the individual. For this reason, there are a series of studies developed to use technology in sign language learning. The conducted studies reported the successful results of humanoid robots developed in sign language learning. It has been proven that the developed humanoid robots are successfully performed in sign language learning. There is no developed humanoid robot for sign language learning in Türkiye. On the other hand, the current studies in the literature have been carried out with a limited number of ready-made humanoid robots including Turkish sign language expressions. With the humanoid robot we developed within the scope of the study, some Turkish sign language statements were successfully realized. [ABSTRACT FROM AUTHOR]

Fire is a natural disaster that causes ecological, social and economic damage. With global warming and the widespread use of explosive/flammable chemicals, fires have become one of the most important problems for humanity. Early fire detection is critical to minimize destruction. For this reason, studies on fire detection through images have begun to be carried out in order to detect fires early. In recent studies for fire detection from images deep learning algorithms are generally used. These studies focus on the analysis of images taken from cameras with models running on powerful servers. With the developments in mobile devices and internet of things, images can analyzed on edge devices. In the study fire and smoke detection model which requires low processing power has been developed to enable the images to be analyzed on mobile device without transferring them to server. The MobileNet convolutional neural network was revised, the last 3 layers were removed and replaced with a flattening layer and a dense layer consisting of two nodes. Fire and smoke detection model, the method with the highest accuracy was determined among the models developed using revised MobileNet, original MobileNet, MobileNetV2, EfficientNetB0, ShuffleNet, NASNetMobile and PeleeNet convolutional neural networks. To train and test the models, 80% of the dataset consisting of 43,355 images was used for training and 20% for testing. According to test results, highest accuracy rate of 98.37% was achieved in the revised MobileNet. [ABSTRACT FROM AUTHOR]

In recent years, Covid-19, which has become a reality of our lives and has become a pandemic for the whole world, increases the rate of infection and even variants of it begin to appear if the necessary precautions are not strictly followed. As the measures published by WHO and necessary to be taken are taken, the fight against the disease may become easier. Although it is difficult to comply with the measures, if care is taken to comply, the disease is either milder or the disease is not easily caught. One of the most important of these measures is to pay attention to the use of masks in crowded areas. After the importance of mask use was supported by research, inspections for the use of masks in crowded places such as some shopping malls, health institutions and schools began. However, since it is difficult for a human to perform these inspections, mask detection studies have begun to be carried out with deep learning methods, which are frequently used today. In this thesis, it is aimed to perform mask detection using transfer learning based models. DenseNet121, EfficientNetV2M, NasNetMobile, InceptionV3, VGG19 and InceptionResNetV2 deep learning models were used with a total of 906 images with the data set available on the Kaggle website. As a result of the experimental evaluations, it was seen that the best success rate was obtained with the NasNetMobile model, with 99.35% accuracy, 99% precision, 99% recall and 99% f1 scores. [ABSTRACT FROM AUTHOR]

Alzheimer's disease is the most common type of dementia in the world, and the treatment methods currently used are aimed only at preventing the progression of the disease. Brain tissue volume changes due to Alzheimer's disease (AD). With the help of tensor-based morphometry (TBM), changes in brain tissues caused by the disease can be monitored. This study aimed to develop an effective method to differentiate between AD patients and the Cognitive Normal (CN) group subjects. TBM, or small local volume differences, are adopted as classification features. Axial brain image slices with 5-pixel intervals covering the hippocampus and temporal lobe from 3D TBM morphometric images belonging to the AD/CN class were recorded in 2D. Then, 60% of the dataset obtained from each clinical group (AD; CN) was allocated as training, 20% as validation, and 20% as test datasets (training: 480; validation: 120; testing: 120). The model performed AD/CN estimation with validation (92.5%) and testing (89%) accuracy values. The results show that TBM obtained from slices covering the hippocampus and temporal lobe with deep learning can be applied with high accuracy in the diagnosis of AD. [ABSTRACT FROM AUTHOR]

The developments in semiconductor circuit manufacturing technologies make the wafer production processes on which these structures are located more complex and sensitive. Various factors such as environmental conditions associated with manufacturing and material quality directly affect the wafer yield, that is, the amount of space on the wafer that can be considered perfect. Possible patterns showing defective areas on a semiconductor wafer are defined as standard. Classifying the defects on the wafer according to these definitions can provide important information to diagnose the causes of problems that occur in the manufacturing processes. In this study, a decision support method has been developed that automatically performs the wafer defect model classification process up to a certain security value. In this method, the basic classification process is performed by a network structure trained with deep learning methods. Wafers classified with an accuracy above the desired safety value are considered to be correctly classified, only the wafers below this value are subject to the inspection of the human operator. By using the method, the total daily inspection time for an average sized wafer manufacturing facility can be reduced to 10% of the time that would apply if the entire inspection was done by the human operator. In addition, subjective evaluations by the human operator can be avoided to a large extent. [ABSTRACT FROM AUTHOR]

Thanks to the opportunities provided by developing technology, there had an increase in the transactions carried out using the internet. This development also led to an increase in data. This situation created the need for new technology for businesses to store, share, control, and manage data securely. One of the current technologies that can be used in this context is the blockchain structure. The blockchain structure is a technology that can be used in many areas, and the most popular usage area today is cryptocurrencies. In this study, it is aimed to estimate Polkadot cryptocurrency, which is one of the essential sub-cryptocurrencies. In the study, the data between 20.08.2020 and 27.02.2023 are used. According to these data, it aimed to estimate the daily average Polkadot value as the output value. Clusters for input values are created in two different ways. In the first input values; number of Polkadot YouTube search, number of Polkadot Google search, and Polkadot volume are used. Unlike the first input values, Ethereum, the leader of the alt cryptocurrencies, is added in the second input value. In this study, which consists of two different input structures, to estimate the daily average values of the Polkadot currency, an estimation study is carried out using multi-layered sensors in artificial neural networks and a long-shortterm memory structure, which is one of the deep learning methods. When the results are examined, it is determined that the values of 4 input sets in the obtained artificial neural networks gave better results with a correlation coefficient of 0.93. [ABSTRACT FROM AUTHOR]

The unregulated and unconscious use of energy sources causes environmental damage. On the other hand the increasing population density, development of industry and technology increase the demand for electrical energy day by day. For this purpose, investments in the energy sector are directed towards renewable energy sources such as wind energy in order to ensure both environment-friendly and supply-demand balance. The amount of energy obtained from wind energy varies depending on regional differences such as wind direction and speed. In this study, a method is proposed to achieve better results in predicting electricity generation from wind energy by capturing the non-linear and non-stationary nature of wind energy using a hybrid approach of deep learning methods, specifically CNN and BLSTM architectures. In the forecasting model, 26280 real-time data measured at hourly frequency are used. In addition, the hyperparameter values used in the model were optimized using the Grid Search algorithm in order to increase the prediction success. The success of the proposed hybrid model is compared with the BLSTM model. As a result, the R² value, which indicates the success rate of the proposed CNN-BLSTM model, was calculated as 0.984. [ABSTRACT FROM AUTHOR]

Blood pressure is the pressure exerted by the blood in the vessels against the vessel walls. If this pressure value is above normal levels, it is known as high blood pressure (HBP) or hypertension (HPT). The diagnosis of this health problem, which negatively affects the quality of life, often causes various damage to organs and can lead to death, is very important. In this study, a method for automatic diagnosis of HPT using ballistocardiography (BCG) signals is proposed. For this, BCG signals are transferred to the time-frequency domain using the continuous wavelet transform filter bank (CWTFB) method. The quality of the conversion has been increased by making some parameter settings in the conversion method used while performing these operations. Then, the obtained images were classified with ResNet18, ResNet50, VGG16 and AlexNet convolutional neural networks and the obtained results were compared. With the proposed method, classification accuracy of 98.92%, 99.34%, 99.22% and 98.07% was obtained for ResNet18, ResNet50, VGG16 and AlexNet architectures, respectively. These high classification results obtained prove that the proposed method can be used for the diagnosis of hypertension. [ABSTRACT FROM AUTHOR]

In this research, we delve deep into the realm of Targeted Social Engineering Email Detection, presenting a novel approach that harnesses the power of Lambda Architecture (LA). Our innovative methodology strategically segments the BERT model into two distinct components: the embedding generator and the classification segment. This segmentation not only optimizes resource consumption but also improves system efficiency, making it a pioneering step in the field. Our empirical findings, derived from a rigorous comparison between the fastText and BERT models, underscore the superior performance of the latter. Specifically, The BERT model has high precision rates for identifying malicious and benign emails, with impressive recall values and F1 scores. Its overall accuracy rate was 0.9988, with a Matthews Correlation Coefficient value of 0.9978. In comparison, the fastText model showed lower precision rates. Leveraging principles reminiscent of the Lambda architecture, our study delves into the performance dynamics of data processing models. The Separated-BERT (Sep-BERT) model emerges as a robust contender, adept at managing both real-time (stream) and large-scale (batch) data processing. Compared to the traditional BERT, Sep-BERT showcased superior efficiency, with reduced memory and CPU consumption across diverse email sizes and ingestion rates. This efficiency, combined with rapid inference times, positions Sep-BERT as a scalable and cost-effective solution, aligning well with the demands of Lambda-inspired architectures. This study marks a significant step forward in the fields of big data and cybersecurity. By introducing a novel methodology and demonstrating its efficacy in detecting targeted social engineering emails, we not only advance the state of knowledge in these domains but also lay a robust foundation for future research endeavors, emphasizing the transformative potential of integrating advanced big data frameworks with machine learning models. [ABSTRACT FROM AUTHOR]

The restrictions and closures experienced during the COVID-19 pandemic period have affected the global supply chains greatly. The logistics sector is among the most affected sectors from this process. For this reason, accurate and fast estimation of logistics demand is important for effective resource planning. In this study, the aim is to predict the demand accurately in a third-party logistics company during the COVID-19 pandemic period. The shipment data of a logistics company between June 2020 and December 2020 were examined, and the prediction problem was considered as univariate time series. In the scope of the study, a deep learning-based demand forecasting model is proposed. In the proposed prediction model, convolutional neural network (CNN) and long short-term memory (LSTM) network are integrated. CNN provides feature extraction, LSTM captures long-term dependencies, and the proposed model is called hybrid CNN-LSTM. The prediction performance of the hybrid CNN-LSTM was evaluated by comparing it with the classical prediction approaches as well as machine learning and deep learning approaches. The parameter values of all forecasting methods were determined by experimental studies. According to the experimental results, the proposed hybrid CNN-LSTM method showed higher performance than the other methods. The proposed approach generates input to workforce and resource planning activities by providing accurate estimation of logistics demand. [ABSTRACT FROM AUTHOR]

The new type of coronavirus disease (COVID-19), which has emerged in recent years, has become a serious disease that threatens health worldwide. COVID-19, which can be transmitted very quickly and with serious increases in death, has paved the way for many concerns. With the spread of the epidemic to a universal dimension, many studies have been carried out for the early diagnosis of this disease. With early diagnosis, both fatal cases are prevented and the planning of the epidemic can be easier. The fact that X-ışını images are much more advantageous than other imaging techniques in terms of time and applicability, and also that they are economical, has led to the focus of early diagnosis-based applications and methods on these images. Deep learning approaches have had a great impact in the diagnosis of COVID-19, as in the diagnosis of many diseases. In this study, we propose a diagnostic system based on the transformer method, which is the most up-to-date and much more popular architecture than previous techniques of deep learning such as CNN-based approaches. This method includes an approach based on vision transformer models and a more effective diagnosis of COVID-19 disease on a new dataset, the COVID-QU-Ex dataset. In experimental studies, it has been observed that vision transformer models are more successful than CNN models. In addition, the ViT-L16 model showed a much higher performance compared to similar studies in the literature, providing test accuracy and F1- score of over 96%. [ABSTRACT FROM AUTHOR]

Rice is a primary food source and is one of the rare plants commonly used in industry. Early diagnosis of leaf diseases in rice is crucial to minimize crop damage. Recently, deep learningbased computer-aided systems have gained importance in the agricultural sector and have played an effective role in various applications. These systems not only help with early disease diagnosis but also serve as a secondary aid to those working in agriculture. This study aims to investigate the effectiveness of deep learning methods in the early diagnosis of diseases in rice leaves. To this end, the most popular convolutional neural networks (CNNs), such as VGG, ResNet, DenseNet, EfficientNet, Inception and Xception, were evaluated on the public Paddy Doctor dataset. Current techniques, such as data preprocessing, data augmentation, hyperparameter optimization, and transfer learning, were applied to each model to increase the diagnostic accuracy of the test set. Additionally, the success of each model in diagnosing diseases in rice leaves was compared in detail to other models. The experimental results showed that the EfficientNetv2_Small model performed better than all other models with a test accuracy of 98.01% and F1-score of 97.99%, outperforming other studies in the literature. This study demonstrates that CNN architectures perform well and can effectively assist agricultural engineers and farmers in the early diagnosis of such diseases. [ABSTRACT FROM AUTHOR]

Sign language is a nonverbal communication tool used by deaf and dumb individuals to convey their feelings, thoughts and social identities to their environment. Sign language has a key role in communication between deaf and dumb individuals and the rest of the society. Many sign language recognition systems have been developed with the increase in human-computer interaction and the fact that sign language is not widely known among normal people. In this study, a new number-based data set for Turkish sign language is proposed for the first time in the literature. The most up-to-date deep learning approaches have been applied to the proposed data set in order to classify Turkish sign language autonomously and to enable computer-based communication of people who have difficulties in this regard. The most up-to-date and popular architectures such as CNN-based VGG, ResNet, MobileNet, DenseNet and EfficientNet were used in the study. In experimental studies, it has been observed that the ResNet152 model performs better than other models with 98.76% accuracy, 98.85% precision, 98.81% sensitivity and 98.80% F1-score. Additionally, the other models used in experimental studies all show a success rate above 90%, supporting the effectiveness of the proposed data set. This shows that CNN models can successfully detect Turkish sign language. [ABSTRACT FROM AUTHOR]

Mosaics, one of the oldest known works of art, have been developed and used by many different civilizations throughout history. Destruction is frequently encountered in mosaic works that have survived from the past to the present. Artifacts can be damaged due to the natural conditions, the negative effects of people or the nature of the objects. The necessity of repairing the damage in these artifacts and reaching their original appearance is the basic need of mosaic artifacts as in all historical artifacts. Image inpainting problem is a current problem that is tried to be solved with different techniques in the literature. In this study, the results of the image inpainting problem on the mosaic data set with deep learning-based methods were examined. Image inpainting architecture is used with contextual attention to correct missing regions in the mosaic image. Comparative results of this architecture with different adversarial generator network architectures were examined using the same data set. Model was retrained with the mosaic dataset using learning transfer. It was observed that the index of structural similarity between the original image and the repaired image in the tested mosaic samples ranged from 0.92 - 0.95 in lightly damaged images and between 0.72 - 0.89 in heavily damaged images, according to the damage ratio. With the implemented image inpainting model, high success was achieved in image inpainting in mosaic paintings with little damage. [ABSTRACT FROM AUTHOR]

The main aim of this study is to examine the secondary school students' self-regulation skills, social studies learning approaches and perceived academic participation. The participants of the research consists of 1088 secondary school students selected by cluster sampling in the Central District of Uşak. "Perceived Self-regulation Scale", "Social Studies Learning Approaches Scale", "Perceived Parent and Teacher Academic Engagement Scale" and Personal Information Form were used as data collection tools in the study in which the relational screening model was used. Independent samples t-test, ANOVA and multiple linear regression analysis were used. According to the results of the research, secondary school students' self-regulation skills are at the "often" level, while their social studies learning approaches and parent-teacher academic participation are at the "agree" level. In addition, parent and teacher academic participation and deep learning approaches significantly predict the self-regulation skills. [ABSTRACT FROM AUTHOR]

The COVID-19 virus, which caused the declaration of a worldwide pandemic in 2019, caused serious health problems and even death due to the rapid progression of the disease. Hence, early diagnosis of the disease has been of great importance to prevent the emergence of these adverse conditions. Medical imaging methods, such as X-Ray chest images, Computed Tomography (CT), etc. have played an important role in the detection. Deep learning methods have proven their superior image analysis ability. In this study, it is proposed to use the Vision Transformer (VIT), which has recently been popular in computer vision applications, for Covid-19 detection using X-ray chest images. The effectiveness of the VIT architecture is compared with a popular Convolutional Neural Network architecture, Resnet50 through various criteria, as well as VGG16 and InceptionV3 architectures, which were recorded to be successful in multi-class datasets of COVID-19. As a result of the experiments, it has been observed that the VIT architecture outperforms the Resnet50 and VGG16 architectures trained with the transfer learning method, and it has been observed that it achieves head-to-head results with InceptionV3, and it has been concluded that it can be used as an alternative decision support system for COVID-19 detection. [ABSTRACT FROM AUTHOR]

The method of obtaining a single high dynamic range (HDR) image from two or more low dynamic range (LDR) images of the same scene is called multi-exposure image fusion (MEF). In this study, a new MEF method using convolutional neural network (CNN) from deep learning (DL) models is proposed. In the proposed method, firstly, the fusion map (fmap) was computed from source LDR images using the CNN model. In order to eliminate the saw-tooth effect in the fused images, weighting was performed on the fmap. Then, a well-exposed fused image was constructed using the weighted fmap. After that, the proposed method was applied to the MEF datasets, which are widely used in the literature, and the fused images obtained were evaluated using well-known quality metrics. The developed technique and other well-known MEF techniques are compared in terms of quantitative and visual evaluation. The results obtained show the feasibility of the proposed technique. [ABSTRACT FROM AUTHOR]

Natural language processing, text mining and deep learning methods are used to obtain the desired information from textual data in many different sectors. Recently increasing number of R&D projects and differentiated project activity areas, as well as the determination of research areas to which R & D projects belong and the determination of reviewer suitable for these research areas, the support processes of projects may be negatively affected. In this article, in order to classify R&D projects, the data in the database where the study was carried out was primarily cleared and in order to classify the features by using the word representation method and automatic feature learning approach "Word2Vec", which is one of the natural language techniques, Convolutional Neural Network (CNN) models have been tried to be created. The evaluation results obtained from experimental studies on the data set of R&D projects and articles with R&D project submitted in the last ten years content selected from the TUBITAK Dergipark site were compared with other classic algorithms and showed that CNNs with Word2Vec models in particular produce more effective results with many performance parameters. [ABSTRACT FROM AUTHOR]

Patents are documents that help protect innovations in information technologies and grant special rights to the creator of these innovations for a certain period of time. While these rights give the patent owner the right to use the innovation commercially, they prevent others from using the innovation without permission. Radical innovations and ground-breaking technological advances are derived from technical information contained in existing patents. Using an automatic classification system, patents assigned to the technical class to which they belong can pave the way for researchers and provide an environment in which they can create new inventions. This study presents an automatic patent classification analysis using the BERT algorithm. Hyperparameter analyses are also preferred in this study in order to achieve more successful prediction accuracy in automatic patent classification problems. The obtained results were at a level that competed with those in the literature. An accuracy of 58% was achieved at the subclass level. [ABSTRACT FROM AUTHOR]

To fully utilize the potential of solar energy and effectively operate solar energy systems, it is vital to know solar radiation values. Modeling solar radiation data with high variability is a complex problem, and nonlinear methods are needed. This study proposes a hybrid model using CNN and LSTM architectures for solar radiation prediction. The performance and applicability of the proposed model are examined by comparing it with ARIMA statistical method and different machine learning methods such as Random Forest, Decision Tree and K-Nearest Neighbor. The study used hourly solar radiation values measured with a pyranometer on the Afyon Kocatepe University campus. As a result of the experimental studies, the CNNLSTM model is the most successful model in terms of RMSE, MAE, MAPE and r² evaluation criteria, achieving 16.92%, 17.25%, 17.24%, 22.89% more successful results than the Decision Trees model, which produced the most unsuccessful predictions, respectively. [ABSTRACT FROM AUTHOR]

The impact of Covid 19 cases is increasing worldwide due to not complying with social distancing and mask-wearing rules in congested areas such as hospitals, schools, and malls where people have to be together. Although the authorities have taken various precautions to prevent not wearing masks, it is challenging to inspect masks in crowded areas. People who do not wear masks can be unnoticed by visual inspections, which is a critical factor in the increase of the epidemic. This study aims to create an Artificial Intelligence (AI) based mask inspection system with the YOLO V7 deep learning method to ensure that overcrowded public areas are protected from the Covid-19 epidemic. [ABSTRACT FROM AUTHOR]

With the development of technology, artificial intelligence-based applications are used for support in many areas. The health sector is one of the areas where such applications are widely used. The increase in knowledge in the health sector due to technological development has led to the need for expertise in radiological evaluation. Considering the intensive working hours, the inaccessibility of specialists from every branch in health institutions and the importance of early diagnosis, especially in emergency pathologies, the importance of the need for applications that will support physicians in the diagnosis process is understood. In the scope of the study, Visual Geometry Group (VGG), Residual Neural Network (ResNet) and EfficientNet architectures, which are among the current deep learning methods, were applied to PhysioNet, a recent dataset, in order to detect brain hemorrhages using Computed Tomography (CT) images. The models were compared among themselves and with existing studies in the literature using accuracy, precision, recall and F1 score metrics. With this study, the importance of choosing the appropriate model for the dataset has been demonstrated through current models. The success of the EfficientNet-B2 model was higher than both the studies in the literature and the models evaluated within the scope of the article. The results show that current deep-learning models have the potential to help in the diagnosis of an intracranial hemorrhage. The study is essential in terms of early diagnosis of intracranial hemorrhage by at least alerting general practitioners, who bear the burden of emergency services, to the presence of intracranial hemorrhage and ensuring that the bleeding condition is not overlooked. [ABSTRACT FROM AUTHOR]

The single image super resolution problem has been extensively studied in the literature using various deep learning-based techniques. Super resolution based on deep convolutional networks has become a rapidly growing area of interest with many practical applications. However, the first studies based on deep learning were based on convolutional neural networks and peak signal-to-noise ratio (PSNR) oriented. Thanks to the models developed based on generative adversarial networks (GAN) in recent years, it has been determined as the main objective to increase the visual quality; however, this is not seen when the image quality metrics are examined. In this study, both mean square error and perceptual loss values were used for the network loss used during the training of the network. Also, the combination of three different training datasets was used as a new training dataset. As a result of these factors, both the visual quality has been increased and a significant increase has been achieved in the image quality metric values. In addition, batch normalization layers are not included in the network architecture and the training speed of the deep network architecture is increased by using the skip connection technique. The success performance of the proposed model was compared with the state-of-the-art models in the literature. Here, the peak signal to noise ratio (PSNR) and structural similarity index measure (SSIM) values were calculated and evaluated separately for three different test data sets commonly used in the literature. When the obtained results are evaluated, it is seen that the proposed model is more successful than other models and generates higher quality images. When all the findings are evaluated, it is seen that the proposed model is a more efficient model in terms of both success and training speed compared to state-of-the-art models. [ABSTRACT FROM AUTHOR]