Search

The internet holds a significant role in all aspects of our lives, and its importance continues to grow each day. Therefore, the usability of the Internet holds great significance. Low data quality and disinformation severely impact the usability of the internet. Consequently, people face challenges in obtaining accurate and clear information. In the present day, websites predominantly feature image-based content like pictures and videos, as opposed to text-based content. The classification of such content holds immense importance for search engines. As a result, the classification of web pages stands as a crucial research area for scholars. This study focuses on the classification of image-based web pages. A deep learning-based approach is proposed to categorize web pages into four main groups: tourism, machinery, music, and sports. The suggested method yielded the most favourable outcomes when utilizing the Stochastic Gradient Descent (SGD) optimization method, achieving an accuracy of 0.9737, a recall of0.9474, an Fl score of 0.9474, and an Area Under the ROC Curve (AUC) value of 0.9649. Furthermore, the utilization of Deep Learning (DL) led to achieving the most advanced results in web page classification within the existing literature, particularly on the WebScreenshots dataset. [ABSTRACT FROM AUTHOR]

In the study, the rates of impact energy absorption of Acrylonitrile Butadiene Styrene (ABS) fractures produced by the Fused Deposition Modeling (FDM) method were examined. Charpy impact test results were determined using layer thickness, printing speed, support angle, build orientation, notch type, and unfill type. Box-behnken experimental design design in the study. Notch impact samples are produced on an ABS Three-dimensional Printer (3DP). Then, charpy impact tests were performed on the impact test device. Data were evaluated using the Minitab 21 program. Later, Deep Learning (DL) and Extreme Learning Machines (ELM) file models were created based on this development. The best results were obtained as 0.844 kJ/m2 with a layer thickness of 0.09 mm. At 60 mm/s printing speed and 30° support angle, the impact energy absorption is 0.803 kJ/m2. The extinction edge of the highest impact energy is 0.841 kj/m2. The most effective impact absorption was obtained as 0.827 kJ/m2 in the U notch type. In the full infill type, impact energy absorption is obtained as 0.777 kJ/m2. In DL, man is the programming and tanh is the activation function. DL, Mean Squared Error (MSE) value was calculated as 0.000923, r2 was calculated as 0.97427. In ELM, the activation function is sigmoidal at the input and linear at the output. [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]

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]

One of the most important technological developments in the field of medicine is Computed Tomography and Magnetic Resonance imaging techniques. This technique allows the size and shape of tumor areas in body tissues to be determined, making it easier for specialists to determine the type of tumor as well as whether it is benign or malignant. Various deep learning-based computer software have been developed to accurately detect tumor areas in tissue. Due to the lack of image data used in deep learning studies, a limitation naturally arises in studies in this field. In order to eliminate the lack of image data in these studies, image augmentation can be performed using deep learning methods as well as data augmentation methods using various image processing techniques. In this study, Generative Adversarial Networks, a deep learning technique, were employed to duplicate brain MR images and generate synthetic images. After the resulting MR images were made usable by undergoing various pre-processing, similarity rates to real images were calculated using metrics such as Peak Signal-to-Noise Ratio, Structural similarity index and Mean Square Error, and by looking at these rates, realistic images were added to the data set and the data set was expanded. [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]

In the face of the rapidly increasing world population today, researchers have turned to studies aiming to use existing resources more effectively and efficiently, while also exploring new resources to meet the increasing demands, such as raw materials and nutrients. Hydroponic agriculture has been gaining popularity day by day as one of the alternative methods to fulfill crucial human needs. Its characteristics, including resistance to weather conditions, indoor applicability, and vertical orientation, set hydroponic agriculture apart from traditional methods. The absence of soil in this agricultural approach necessitates more observation and supervision. In this study, the use of machine learning was investigated to overcome the observation and surveillance processes that must be done continuously for the healthy growth of plants. Firstly, we developed a hydroponic prototype to assess this goal. With the developed hydroponic prototype, plant water and wastewater data to be used in the evaluation of the growth and development of plants were obtained by sensors. The data obtained through Arduino was stored in the database and used in the training processes of machine learning algorithms. Our experimental study, which utilized five machine learning and deep learning methods, demonstrated a significant increase in hydroponic agricultural efficiency. Notably, deep learning outperformed other methods with a 99.7% success rate. In conclusion, our study shows that machine learning can be used effectively in hydroponic agriculture by providing observation and surveillance of plants. [ABSTRACT FROM AUTHOR]

Diabetes mellitus has been on the rise recently. Diabetes can adversely affect different organs in the human body, along with blood flow. Diabetic retinopia is a type of diabetes that causes vision loss as a result of sugar destroying the vessels in the retina of the eye. Diabetic retinopia is one of the most important causes of blindness. Optic disc segmentation can help specialists in identifying these diseases. However, there are some obstacles, such as different brightness states on the images. Vessel noise can reduce segmentation success. This is because blood vessels can cut the edges of the optic disc. The aim in this study is to remove blood vessels from the images and then successfully segment the optic disc. Within the scope of the study, a new approach named LinkNetRCB7 based on LinkNet was developed. Successful results were obtained in the trainings made with LinkNetRCB7 data sets. The blood vessels and optic disc segmentation accuracy was 98.50% in the STARE dataset and 98.85% in the DRISHTI GS. A decision support system including these stages has been proposed within the scope of the study. There is no decision support system that includes deep learning and image processing algorithms for diabetic retinopia in the current literature. [ABSTRACT FROM AUTHOR]

Analysis of microscopic images is a reliable laboratory method that provides useful information for disease diagnosis in medical field. Although advanced technological devices can help with blood disease diagnosis, a microscopic blood smear examination is required for a definitive diagnosis. Nowadays, technicians in many laboratories use microscope to detect anomalies in cells (defects in the cell, parasites, low or excess cell count, etc.). Experts' detection of anomalies provides critical information for disease diagnosis. For the expert, analysing microscopic images is a time-consuming and error-prone procedure. As a result, in this study, a method for accelerating expert examination and automatically detecting cells is presented. The importance of basic blood cell segmentation and classification is emphasized. PBC data set blood smear images were used as the data set. The system was built using the Mask R-CNN architecture, which is a region-based convolutional neural network. Mask R-CNN was tested using various spine structures. The sample segmentation feature in the Mask R-CNN algorithm was used to determine the segmentation of blood cells obtained from images, and the error rates were minimized as a result of the tests. The study focused on detecting eight classes but could be improved by enriching it with further classes and using blood cell images from different angles and better segmentation. [ABSTRACT FROM AUTHOR]

Nowadays, the successful transmission of data between end users, data protection and confidentiality are important issues. In our study, a system that ensures the transmission of bitwise encrypted images in pixel color matrix information based on the Lehmer algorithm for the generation of random keys over an end-to-end reliable communication channel and guarantees accuracy validation has been created. Using the deep learning model, metadata is created with the properties of the detected objects in the image scene. In the study, structured data is created in Base64 encoding format after encryption is done by the sender end of the message using the Lehmer algorithm. This message is transmitted end-to-end to the receiving party over the secure transmission channel. Finally, the encoded version of the image and the metadata information are extracted from the structured data exchange and transfer file containing the detection information of the objects in the image scene in the message packet reaching the receiver end, and also, decryption is performed. On the receiver end, object detection results of the deep learning model are compared with the decrypted information obtained from this file, and an objective verification of the accuracy of the message is also performed. Considering the average values in the experiments, the average values with the data obtained from three different data sets are for the Number of Pixels Change Rate (NPCR) 99.61%, the Unified Average Changing Intensity (UACI) 14.70%, and also, the maximum average entropy value is obtained 7.9999 for encrypted images. In our study, discussions and evaluations based on scientific findings are included. [ABSTRACT FROM AUTHOR]

Sleep disorders have high prevalence and cause various health problems. For the diagnostics of these disorders and assessment of the sleep quality, many physiological data are collected using polysomnogram (PSG) method. The most important PSG data is the EEG recorded from the brain during sleep. Analysis of hours of sleep EEG data by experts is an onerous task which requires high attention. Recently, many automatic sleep staging classifiers using EEG are developed in order to prevent human error, and to provide a quick objective analysis. They use machine learning techniques and predict the sleep stage of each EEG epoch. Compared to traditional machine learning, deep learning which requires no hand-crafted feature extraction was able to classify sleep stages better. 1D Convolutional Neural Networks (CNN) are the main methods used in automatic sleep staging recently. In this research a simple 2D-CNN based automatic sleep staging feasibility is investigated. It has been found that a 2D CNN can classify the sleep stages by accuracy of 92.55% and with a Cohen's kappa of 0.82. [ABSTRACT FROM AUTHOR]

With the intensive work done, deep learning finds many useful areas. However, obtaining a sufficient amount of data required by deep learning is not always an easy task. To overcome this difficulty, deep network trainers prefer to develop their datasets by using a set of algorithms. With the increased amount of data, deep networks can be trained more successfully. Data augmentation (DA) is one of the most widely used methods of increasing the amount of data. With DA, the number of sounds and images that a convolutional neural network (CNN) can classify can be increased. In this study, the number of images belonging to 6 classes that do not have enough images to train the CNN successfully enough was increased by DA methods. First, the amount of data was increased by applying three different DA methods separately and all three together. The original dataset and created datasets in which DA methods were used are used to train 15 CNNs with different parameters. Then, their effects on CNN have been investigated. As a result, a success increase of over 5% was observed by increasing the data. [ABSTRACT FROM AUTHOR]

Early diagnosis of COVID-19 disease becomes possible with the enhancements on feature learning and advanced pre-processing stages for classification of chest X-ray images using deep learning. Besides, high-performance models have been developed by many researchers due to the popularity of Deep Learning. In this study, chest X-ray images were pre-processed using Contrast Limited Adaptive Histogram Equalization (CLAHE) before the classification with particular popular transfer learning approaches in deep learning architectures including AlexNet, MobileNet, VGG16, and DarkNet19. The originality of the paper is pre-processing the images using CLAHE to obtain more significant representations of airways and pathologies instead of training with raw chest X-ray images. The best CLAHE parameters were determined considering the results of various trials at a specified range. The other superior contribution of the proposal is using a large-scale dataset, which is comprised of 3500 healthy and 3615 chest x-rays with COVID-19. The CLAHE-based transfer learning proposal achieved an accuracy rate of 95.878% as the most successful binary classification result for COVID-19 and healthy using VGG16 model and CLAHE parameters including disk value of 56, clip-limit of 0.2. [ABSTRACT FROM AUTHOR]

Nowadays, with the increase in the number of employees in the enterprises and the proportional workload, different occupational accidents frequently occur. Examples of this are slippery floors, falling materials, harmful substances/gas leaks, improper use of protective clothing and equipment or not using them at all. Identifying these dangers and taking the necessary precautions are very important for both worker safety and employers. The most common accident among these dangerous situations is the accidents that occur as a result of slipping or falling. Such accidents are usually caused by a foreign liquid/substance on the work surface, the worker's inability to balance himself, or surface inequalities. With this study, an IoT and 1D CNN deep learning-based system has been developed to detect accidents such as falling, slipping and balance disorders to inform relevant health institutions. The developed 1D CNN-based system detected work accidents caused by falls with 100% accuracy. With the results obtained from this study, it is aimed to make improvements for the prevention of these accidents. [ABSTRACT FROM AUTHOR]

In this paper, the approaches developed for content management and automatic content control of the job postings on the İşin Olsun website, which is a blue-collar job search and finding platform, will be explained. For this purpose, experiments were carried out with supervised machine learning models and Bert transformer architecture methods, and the results were observed. At the end of the study, a system was developed in which the contents of the job texts are automatically classified using Bert for sequence classification, and this system was integrated into the job search platform. Starting from labeling the job texts in two different classes as appropriate or unsuitable content by the controllers, from the preparation stages of the dataset to the integration of the classification model into the system, the studies are summarized in this paper. [ABSTRACT FROM AUTHOR]

Sonar is used to determine the size, distance, direction, and other object features using sound waves. It is widely used in submarine mining, oil exploration, submarine mapping, tracking fish shoals, and mine detection. In Machine Learning (ML) research, feature extraction, selection, algorithm selection, and hyper-parameter optimization, which should be used to identify and classify sonar signals, are seen as scientific problems studied for many years. In this study, instead of commonly used ML algorithms and feature extraction processes, three different mathematical transformations were suggested to classify the underwater objects as an innovative approach. This novel approach applied on a data set in time-series format, data has been transformed from onedimensional data to a two-dimensional format and a simple channel merging technique was applied to create a new image joining these images. The methods' performance was measured by the classification results of mines and rocks using deep learning algorithms on the sonar dataset. Moreover, the performance results obtained with deep learning, compared with the classical algorithms. Finally, confronted with other studies in the literature, it has been seen that the proposed time-series data-to-image transformation with a channel-merging approach eliminates the need for feature extraction and achieves superior results against the others. [ABSTRACT FROM AUTHOR]