Volkan Kaya
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Volkan Kaya has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Ahmet Baran
Department of Computer Engineering, Erzincan Binali Yıldırım University, Erzincan 24100, Turkey
1 shared publication - Ahmet Baran
Department of Computer Engineering, Faculty of Engineering and Architecture, Erzincan Binali Yıldırım University, Türkiye
1 shared publication - Büşra Er
Yıldırım University Erzincan-Türkiye Erzincan Binali Yıldırım University Erzincan-Türkiye
1 shared publication - Canan Tastimur
ERZİNCAN ÜNİVERSİTESİ, MÜHENDİSLİK FAKÜLTESİ
1 shared publication - Erdal Karavas
Department of Radiology, Bandırma Onyedi Eylul University School of Medicine, Balıkesir, Türkiye
1 shared publication İsmail Akgül
1 shared publication
Latest Publications
- Open Access
- 24 March 2026
- Published by Ankara University Faculty of Agriculture in Tarım Bilimleri Dergisi
- Vol. 32 (2) , 474-497
LAViT: Class-Aware Vision Transformer with Learnable Attention for Pest Recognition in Agricultural Fields
Insect pests pose a significant threat to agricultural productivity, making early and accurate identification essential for effective pest management. This study proposes a novel deep learning-based classification framework for multi-class pest recognition from field images. The proposed approach enhances discriminative region representation by integrating a patch embedding module, a Vision Transformer backbone, and a learnable spatial attention mask. This hybrid design enables the model to focus on critical visual cues without requiring segmentation-based preprocessing. The attention mask, learned via convolutional layers, is pooled and directly applied to Transformer-encoded patch tokens to refine spatial feature emphasis. Positional embeddings are further employed to preserve spatial context within the tokenized image representation. Experimental evaluations conducted on publicly available pest datasets with 5, 9, and 12 classes demonstrate the effectiveness and robustness of the proposed framework. The model achieves accuracies of 99.67%, 99.52%, and 97.00% on the Pest5, Pest9, and Pest12 datasets, respectively, indicating strong generalization across varying classification complexities. To enhance model transparency and reliability, visual interpretability is provided through Grad-CAM and attention heatmap visualizations that reveal the model’s focus regions. Additionally, t-SNE-based feature visualization illustrates clear separability in the learned embedding space. The proposed framework shows strong potential for practical deployment in smart agriculture and precision pest monitoring systems.
- 24 March 2026
- Published by Ankara University Faculty of Agriculture in Tarım Bilimleri Dergisi
- Vol. 32 (2) , 474-497
Open AccessLAViT: Class-Aware Vision Transformer with Learnable Attention for Pest Recognition in Agricultural Fields
Insect pests pose a significant threat to agricultural productivity, making early and accurate identification essential for effective pest management. This study proposes a novel deep learning-based classification framework for multi-class pest recognition from field images. The proposed approach enhances discriminative region representation by integrating a patch embedding module, a Vision Transformer backbone, and a learnable spatial attention mask. This hybrid design enables the model to focus on critical visual cues without requiring segmentation-based preprocessing. The attention mask, learned via convolutional layers, is pooled and directly applied to Transformer-encoded patch tokens to refine spatial feature emphasis. Positional embeddings are further employed to preserve spatial context within the tokenized image representation. Experimental evaluations conducted on publicly available pest datasets with 5, 9, and 12 classes demonstrate the effectiveness and robustness of the proposed framework. The model achieves accuracies of 99.67%, 99.52%, and 97.00% on the Pest5, Pest9, and Pest12 datasets, respectively, indicating strong generalization across varying classification complexities. To enhance model transparency and reliability, visual interpretability is provided through Grad-CAM and attention heatmap visualizations that reveal the model’s focus regions. Additionally, t-SNE-based feature visualization illustrates clear separability in the learned embedding space. The proposed framework shows strong potential for practical deployment in smart agriculture and precision pest monitoring systems.
- 25 July 2025
- Published by NWSA Academic Journals in NWSA-Engineering Sciences
- Vol. 20 (3) , 79-93
A COMPARATIVE STUDY OF CNN AND TRANSFORMER-BASED DEEP LEARNING MODELS FOR TEA LEAF DISEASE RECOGNITION
This study presents a comparative analysis of six deep learning models for the automatic classification of eight different disease categories found in tea leaves. The dataset used in the study was divided into three parts with 70% training, 15% validation, and 15% testing ratios. As part of the experimental evaluation, five convolutional neural network (CNN) based architectures (ResNet50, DenseNet121, EfficientNet-B0, MobileNetV3-Large, and ConvNeXt-Tiny) and one Transformer-based model (Vision Transformer, ViT-Small) were tested using the same training strategies. The models were trained using a transfer learning and finetuning approach; performance metrics were reported based on accuracy, precision, recall, and F1-score values. In addition, the number of parameters and the prediction time per image were calculated for each model. Experimental results show that the DenseNet121 model achieved the highest success rate in the validation dataset, while the ConvNeXt-Tiny architecture achieved the highest accuracy and F1-score values in the standalone test dataset. The findings indicate that modern CNN-based architectures offer high generalization capabilities in the classification of tea leaf diseases. The results obtained serve as a comparative reference for future studies in the field of agricultural image analysis.
- 25 July 2025
- Published by NWSA Academic Journals in NWSA-Engineering Sciences
- Vol. 20 (3) , 79-93
A COMPARATIVE STUDY OF CNN AND TRANSFORMER-BASED DEEP LEARNING MODELS FOR TEA LEAF DISEASE RECOGNITION
This study presents a comparative analysis of six deep learning models for the automatic classification of eight different disease categories found in tea leaves. The dataset used in the study was divided into three parts with 70% training, 15% validation, and 15% testing ratios. As part of the experimental evaluation, five convolutional neural network (CNN) based architectures (ResNet50, DenseNet121, EfficientNet-B0, MobileNetV3-Large, and ConvNeXt-Tiny) and one Transformer-based model (Vision Transformer, ViT-Small) were tested using the same training strategies. The models were trained using a transfer learning and finetuning approach; performance metrics were reported based on accuracy, precision, recall, and F1-score values. In addition, the number of parameters and the prediction time per image were calculated for each model. Experimental results show that the DenseNet121 model achieved the highest success rate in the validation dataset, while the ConvNeXt-Tiny architecture achieved the highest accuracy and F1-score values in the standalone test dataset. The findings indicate that modern CNN-based architectures offer high generalization capabilities in the classification of tea leaf diseases. The results obtained serve as a comparative reference for future studies in the field of agricultural image analysis.
- 30 January 2024
- Published by Polish Academy of Sciences Chancellery in Bulletin Of The Polish Academy Of Sciences-Technical Sciences
- Vol. 72 (3) , 149172
A novel artificial intelligence-based hybrid system to improve breast cancer detection using DCE-MRI
The interpretation of breast magnetic resonance imaging (MRI) in the healthcare field depends on the good knowledge and experience of radiologists. Recent developments in artificial intelligence (AI) have shown advances in the field of radiology. However, the desired levels have not been reached in the field of radiology yet. In this study, a novel model structure is proposed to characterize the diagnostic performance of AI technology for individual breast dynamic contrast material–enhanced (DCE) MRI sequences. In the proposed model structure, Inception-v3, EfficientNet-B3, and DenseNet-201 models were used as hybrids together with the Yolo-v3 algorithm to detect breast and cancer regions. In the proposed model, DCE-MRI sequences (T2, ADC, Diffusion, Non-Contrast Fat Non-Suppressed T1, Non-Contrast Fat Suppressed T1, Contrast Fat Suppressed T1, and Subtraction T1) were evaluated separately and validation was made, thus providing a unique perspective. According to the validation results, the model structure with the best performance was determined as Yolo-v3 + DenseNet-201. With this model structure, 92.41% accuracy, 0.5936 loss, 92.44% sensitivity, and 92.44% specificity rates were obtained. In addition, it was determined that the results obtained without using contrast material in the best model were 91.53% accuracy, 0.9646 loss, 92.19% sensitivity, and 92.19% specificity. Therefore, it is predicted that the need for contrast material use can be reduced with the help of this model structure.
- 30 January 2024
- Published by Polish Academy of Sciences Chancellery in Bulletin Of The Polish Academy Of Sciences-Technical Sciences
- Vol. 72 (3) , 149172
A novel artificial intelligence-based hybrid system to improve breast cancer detection using DCE-MRI
The interpretation of breast magnetic resonance imaging (MRI) in the healthcare field depends on the good knowledge and experience of radiologists. Recent developments in artificial intelligence (AI) have shown advances in the field of radiology. However, the desired levels have not been reached in the field of radiology yet. In this study, a novel model structure is proposed to characterize the diagnostic performance of AI technology for individual breast dynamic contrast material–enhanced (DCE) MRI sequences. In the proposed model structure, Inception-v3, EfficientNet-B3, and DenseNet-201 models were used as hybrids together with the Yolo-v3 algorithm to detect breast and cancer regions. In the proposed model, DCE-MRI sequences (T2, ADC, Diffusion, Non-Contrast Fat Non-Suppressed T1, Non-Contrast Fat Suppressed T1, Contrast Fat Suppressed T1, and Subtraction T1) were evaluated separately and validation was made, thus providing a unique perspective. According to the validation results, the model structure with the best performance was determined as Yolo-v3 + DenseNet-201. With this model structure, 92.41% accuracy, 0.5936 loss, 92.44% sensitivity, and 92.44% specificity rates were obtained. In addition, it was determined that the results obtained without using contrast material in the best model were 91.53% accuracy, 0.9646 loss, 92.19% sensitivity, and 92.19% specificity. Therefore, it is predicted that the need for contrast material use can be reduced with the help of this model structure.
- Open Access
- 25 August 2023
- Published by Frontiers Media SA in Frontiers in Environmental Science
- Vol. 11, 1228732
Classification of waste materials with a smart garbage system for sustainable development: a novel model
In today’s conditions, where the human population is increasing, environmental pollution is also increasing around the world. One of the most important causes of environmental pollution is the waste materials in the garbage. Misuse of waste materials causes significant damage to both the environment and human health. With the developing technology, the recyclability of the raw material used in the production of waste materials significantly affects both the raw material needs of the countries and the energy savings. Therefore, many traditional activities are carried out in recycling facilities in order to reuse the waste materials that can be recycled in many countries. At the beginning of these activities is manual waste collection and pre-processing depending on the human workforce. This process poses a serious threat to both the environment and human health. For this reason, there is a need for a smart system that automatically detects and classifies the waste materials in the garbage. In this study, Xception, InceptionResNetV2, MobileNet, DenseNet121 and EfficientNetV2S deep learning methods based on artificial intelligence, which automatically classify the waste materials in the garbage, were used and in addition to these methods, Xception_CutLayer and InceptionResNetV2_CutLayer based on transfer learning techniques were proposed. The proposed methods and artificial intelligence-based deep learning methods were trained and tested with a dataset containing 6 different waste materials. According to the findings obtained as a result of training and testing, a classification success rate of 89.72% with the proposed Xception_CutLayer method and 85.77% with the InceptionResNetV2_CutLayer method, a better success rate was obtained than the other artificial intelligence-based methods discussed in the study.
- 25 August 2023
- Published by Frontiers Media SA in Frontiers in Environmental Science
- Vol. 11, 1228732
Open AccessClassification of waste materials with a smart garbage system for sustainable development: a novel model
In today’s conditions, where the human population is increasing, environmental pollution is also increasing around the world. One of the most important causes of environmental pollution is the waste materials in the garbage. Misuse of waste materials causes significant damage to both the environment and human health. With the developing technology, the recyclability of the raw material used in the production of waste materials significantly affects both the raw material needs of the countries and the energy savings. Therefore, many traditional activities are carried out in recycling facilities in order to reuse the waste materials that can be recycled in many countries. At the beginning of these activities is manual waste collection and pre-processing depending on the human workforce. This process poses a serious threat to both the environment and human health. For this reason, there is a need for a smart system that automatically detects and classifies the waste materials in the garbage. In this study, Xception, InceptionResNetV2, MobileNet, DenseNet121 and EfficientNetV2S deep learning methods based on artificial intelligence, which automatically classify the waste materials in the garbage, were used and in addition to these methods, Xception_CutLayer and InceptionResNetV2_CutLayer based on transfer learning techniques were proposed. The proposed methods and artificial intelligence-based deep learning methods were trained and tested with a dataset containing 6 different waste materials. According to the findings obtained as a result of training and testing, a classification success rate of 89.72% with the proposed Xception_CutLayer method and 85.77% with the InceptionResNetV2_CutLayer method, a better success rate was obtained than the other artificial intelligence-based methods discussed in the study.
- 25 July 2022
- Published by NWSA Academic Journals in NWSA-Engineering Sciences
- Vol. 17 (3) , 35-41
REALIZATION OF A YOLO-V3-BASED APPLICATION FOR THE DETECTION OF TICK-BORNE CASES
- 25 July 2022
- Published by NWSA Academic Journals in NWSA-Engineering Sciences
- Vol. 17 (3) , 35-41
REALIZATION OF A YOLO-V3-BASED APPLICATION FOR THE DETECTION OF TICK-BORNE CASES
- Open Access
- 17 August 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (16) , 7535
Detection and Classification of Different Weapon Types Using Deep Learning
Today, with the increasing number of criminal activities, automatic control systems are becoming the primary need for security forces. In this study, a new model is proposed to detect seven different weapon types using the deep learning method. This model offers a new approach to weapon classification based on the VGGNet architecture. The model is taught how to recognize assault rifles, bazookas, grenades, hunting rifles, knives, pistols, and revolvers. The proposed model is developed using the Keras library on the TensorFlow base. A new model is used to determine the method required to train, create layers, implement the training process, save training in the computer environment, determine the success rate of the training, and test the trained model. In order to train the model network proposed in this study, a new dataset consisting of seven different weapon types is constructed. Using this dataset, the proposed model is compared with the VGG-16, ResNet-50, and ResNet-101 models to determine which provides the best classification results. As a result of the comparison, the proposed model’s success accuracy of 98.40% is shown to be higher than the VGG-16 model with 89.75% success accuracy, the ResNet-50 model with 93.70% success accuracy, and the ResNet-101 model with 83.33% success accuracy.
- 17 August 2021
- Published by MDPI AG in Applied Sciences
- Vol. 11 (16) , 7535
Open AccessDetection and Classification of Different Weapon Types Using Deep Learning
Today, with the increasing number of criminal activities, automatic control systems are becoming the primary need for security forces. In this study, a new model is proposed to detect seven different weapon types using the deep learning method. This model offers a new approach to weapon classification based on the VGGNet architecture. The model is taught how to recognize assault rifles, bazookas, grenades, hunting rifles, knives, pistols, and revolvers. The proposed model is developed using the Keras library on the TensorFlow base. A new model is used to determine the method required to train, create layers, implement the training process, save training in the computer environment, determine the success rate of the training, and test the trained model. In order to train the model network proposed in this study, a new dataset consisting of seven different weapon types is constructed. Using this dataset, the proposed model is compared with the VGG-16, ResNet-50, and ResNet-101 models to determine which provides the best classification results. As a result of the comparison, the proposed model’s success accuracy of 98.40% is shown to be higher than the VGG-16 model with 89.75% success accuracy, the ResNet-50 model with 93.70% success accuracy, and the ResNet-101 model with 83.33% success accuracy.