Search

Deep learning is an important discipline in which human-specific problems are solved with the help of machines with advanced hardware power. It is seen this discipline is widely used in the fields of industry, health, defense industry, and sports. In addition, the use of deep learning in the field of horticulture is an important requirement. With the integration of deep learning into horticulture, to do product classification is very important for increasing productivity and production. In this study, a method using ensemble learning is proposed to improve the accuracy of the classification problem for horticultural data. For this method, a new dataset was created, containing a total of 24421 images and 15 crop classes, independent of data augmentation. In order to train this created data set with the help of the proposed method, a hierarchical structure has been designed in which the output of one model is the input of the other model. A total of 7 pre-trained models were used in the experimental studies of the proposed method. Since this method is in an ensemble structure, it is possible to add or remove pre-trained models from the structure. With the help of experimental studies, a performance analysis of the proposed method, which is compared with the traditional CNN method, has been made. As a result of these analyses, it has been observed that the proposed method works 3% more successfully. [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]

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]

This study is aimed to determine the effect of vehicle deceleration and acceleration on fuel consumption in the bumps which are used to regulate traffic. For this, real-time fuel consumption and speed data are acquired with Arduino from the OBD-II port of the vehicle drived on routes with bumps. Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) deep learning models are developed to predict fuel consumption. The preprocessed data is used to train the models. Hyperparameter optimization is conducted in the developed models. Thus, the number of layers and the units in the layers, the activation functions and the learning rate is specified. The lowest mean square error is obtained as 63? in the validation set. The effects of different speed scenarios on fuel consumption are predicted by using the models. In conclusion, fuel consumption increased between 16.30% and 31.03% during the impact of the bumps, by using the speed and time calculated for the bumps. [ABSTRACT FROM AUTHOR]

Nowadays, trade based on logistics and sea transportation has gained importance. Considering the traffic related to this, the classification and discrimination of ship types are important in terms of transportation and storage costs, and safety. The classification of ships performing different tasks on the sea is handled in this study, and a ship image dataset has been created in order to make a high accuracy ship classification thanks to deep learning methods. It is preferred in our deep learning study in comparison to the classical machine learning method, as the features are semantically richer as the higher level, while expressing content from the dataset. It was created by the acquisition of various ship images thanks to the web scraping method. YOLOv5 and Xception deep learning models were trained to obtain the most appropriate classification performance. As a result of the experiments, an accuracy rate of approximately between 96% and 99% was achieved with both models. Scientific findings and discussion are included in our study as well. [ABSTRACT FROM AUTHOR]

Although air pollution is an important problem for today, reasons such as industrialization, forest fires, exhaust gases, poor quality fuel use confront us with a serious problem that will threaten future generations. Among these reasons, intensive industrialization is one of the most critical factors that play a role in air pollution. Regional industrial development affects air quality in cities. While the amount of some pollutants decreases with the development of the industry, there is an increase in ozone levels. In the coming years, it becomes inevitable to predict air pollution in order not to feel the problems that air pollution will cause more, to manage air quality, and to take precautions against risks. In this study, machine learning and deep learning methods based on time series were applied to predict hourly ozone levels between 2018 and 2021 for the provinces of Kocaeli and Sakarya, where the industry is developed, and Çanakkale, where the industry is not developed much. The applied models were compared using Mean Absolute Error (MAE), Relative Absolute Error (RAE), and R-square (R2) metrics, and it was aimed to determine the most effective method. [ABSTRACT FROM AUTHOR]

Fake news is fabricated news that spread consciously or unconsciously through various communication channels and has no real share. Today, the masses receive most news on digital and social media. In such communication environments, where news can be transferred to the masses quickly, the accuracy of this news can often be abused. News of unknown origin can cause serious problems in societies by making disinformation or misinformation. Especially, fake news exposed to information pollution in the internet environment can show its effect on society very quickly. To prevent such problems in digital environments, an artificial intelligence-based approach that can grasp the accuracy of the news and confirm it quickly is proposed in this study. In addition, a classification analysis was performed using the Natural Language Processing (NLP) method, a sub-branch of artificial intelligence, to determine whether the news was real or false using the dataset that was accessible. The dataset consisted of 6335 news headlines and content. While 3171 of this news is real news; 3164 is fake news. In the analysis of the study, the Long Short Term Memory (LSTM) model was used together with the NLP method and the training of the dataset was carried out with this model. As a result, the overall accuracy success from the training data was 99.83%, and the overall accuracy success from the test data was 91.48%. These results show us that similar studies that we plan to think about in the future have been promising. [ABSTRACT FROM AUTHOR]

Today's manufacturing vision necessitates extracting insights from the data collected in real-time from manufacturing processes. Predicting failures with the predictive analysis of the collected process data and preventing these failures by taking necessary actions before they occur is a key factor in ensuring quality at the desired level, increasing productivity, and reducing costs in production systems. In the literature on predictive analysis of process data, machine learning and deep learning methods have attracted considerable attention, especially in recent years. This study has addressed a multi-class failure classification problem in the plastic extrusion process with a real case study. Classification models have been developed based on Long Short-term Memory (LSTM) as a deep learning method and Multilayer Perceptron (MLP) and Logistic Regression (LR) as machine learning methods to predict the failure categories. In the case study, real data taken from the extrusion process of one of the leading insulation companies operated in Izmir has been used. The final dataset includes actual measurements of seven parameters related to temperature and pressure and failure categories as the target variable. Three failure categories have been identified to define Category 0 (No failure), Category 1 (Filter change), and Category 2 (Feeding failures) states, and coded as 0,1 and 2 in the models, respectively. LSTM, MLP, and LR's performance to predict the failure categories have been evaluated and compared based on accuracy, precision, recall, and F1 Score measures. LSTM is the highest performing among the three methods, with 100% prediction accuracy for each failure category. On the other hand, LR and MLP have achieved considerable and close results except for Category 1. [ABSTRACT FROM AUTHOR]

In the domain of Natural Language Processing (NLP), despite the progress made for some common languages, difficulties persist for many others for the completion of particular NLP tasks. In this scope, the current study aims to explore these challenges by proposing a question answering (QA) system in the Turkish language. In particular, the system will generate the best answers in terms of content and length from questions that are based on a set of documents related to the banking sector. In order to achieve this goal, the system utilizes advanced artificial intelligence algorithms and large data sets. More specifically, BERT algorithm is used for the generation of the language model, followed by a fine-tuning procedure for performing a machine reading for question answering (MRQA) task. In this work, various experiments were conducted using original and translated data sets in an effort to solve the challenges that arise from morphologically complex languages as Turkish. Finally, the system achieved a performance that overall is applicable to a wider range than any other QA system in the Turkish language. The proposed methodology is not only proper to the Turkish language, but can also be adapted to any other language for performing various NLP tasks. [ABSTRACT FROM AUTHOR]