Tayyip Ozcan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Tayyip Ozcan has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Alper Basturk
Erciyes University
4 shared publications - Fatih Horozoglu
Erciyes University Medical Faculty
2 shared publications - Hatice Türkdoğan
Erciyes Üniversitesi, Fen Bilimleri Enstitüsü, Bilgisayar Mühendisliği Anabilimdalı
2 shared publications Ahmet Nusret Toprak
Department of Computer Engineering, Erciyes University Faculty of Engineering, Kayseri, Türkiye
1 shared publication- Ahmet Nusret Toprak
Erciyes Universitesi
1 shared publication Alper Basturk
Erciyes University
1 shared publication
Latest Publications
- 20 July 2026
- Published by Journal of the Faculty of Engineering and Architecture of Gazi University in Gazi Üniversitesi Mühendislik-Mimarlık Fakültesi Dergisi
- Vol. 41 (2) , 1441-1452
Geliştirilmiş YOLO mimarisi kullanılarak optik koherens tomografi (OCT) görüntülerinde gerçek zamanlı hastalık tespiti ve analizi
Retina hastalıkları, gözün retina tabakasını etkileyerek ilerleyici görme kaybına neden olmaktadır. Bu hastalıklar genellikle yavaş ilerlediği için erken dönemde teşhis edilmesi gerekmektedir. Günümüzde uzmanlar retina hastalıklarının teşhisinde Fundus Floressein Anjiografi (FFA) ve Optik Koherens Tomografi (OCT) testlerinden yararlanmaktadır. Ancak, söz konusu testlerin doğru bir şekilde yorumlanabilmesi için alanında uzman hekimlerin değerlendirmesine ihtiyaç duyulmaktadır. Özellikle hasta sayısındaki artış, bu görüntülerin manuel olarak incelenmesini güçleştirmekte; buna ek olarak, kırsal bölgelerde yaşanan uzman hekim eksikliği, tanı süreçlerinde gecikmelere neden olabilmektedir. Bu çalışmada ise Retinal OCT-C8 veri seti kullanılarak 8 farklı retina hastalığı sınıflandırılmıştır. Sınıflandırma sürecinde standart You Only Look Once (YOLO) algoritmasının yanı sıra geliştirilmiş YOLO modelleri de kullanılmıştır. Geliştirilen YOLO modellerinde mimari derinlik katman sayısı artırılarak iyileştirilmiştir. YOLOv5 modelinde ek olarak K-Means ile kümelenmiş anchor boxlar eklenmiştir. Retinal OCT-C8 veri setinde yapılan test işleminde geliştirilmiş YOLOv8 modeli, %98.1 doğruluk, %97.6 kesinlik, %98.6 duyarlılık ve %98 F-Skor değerleri ile bu karşılaştırmadaki en iyi performans gösteren modeldir. Modelin genel performansını değerlendirmek amacıyla farklı veri setleri üzerinde de testler gerçekleştirilmiştir. OCTDL veri seti ile yapılan testlerde %88 doğruluk değeri ve %89, F-Skor değeri, OCTID veri seti ile yapılan testlerde ise %86 doğruluk değeri ve %90 F-Skor değeri elde edilmiştir. Ayrıca bu çalışmada, doktorların retina fotoğraflarını yükleyerek test edebileceği ve hasta bilgilerini düzenleyebileceği bir yazılım geliştirilmiştir. Elde edilen sonuçlar, geliştirilen modelin farklı veri kümeleri üzerinde yüksek doğrulukla çalıştığını ve retina hastalıklarının otomatik teşhisinde etkili bir yöntem sunduğunu ortaya koymaktadır. Ayrıca geliştirilen yazılım ile birlikte, tanı sürecinin kısa sürede tamamlanabileceğini de göstermektedir.
- 20 July 2026
- Published by Journal of the Faculty of Engineering and Architecture of Gazi University in Gazi Üniversitesi Mühendislik-Mimarlık Fakültesi Dergisi
- Vol. 41 (2) , 1441-1452
Geliştirilmiş YOLO mimarisi kullanılarak optik koherens tomografi (OCT) görüntülerinde gerçek zamanlı hastalık tespiti ve analizi
Retina hastalıkları, gözün retina tabakasını etkileyerek ilerleyici görme kaybına neden olmaktadır. Bu hastalıklar genellikle yavaş ilerlediği için erken dönemde teşhis edilmesi gerekmektedir. Günümüzde uzmanlar retina hastalıklarının teşhisinde Fundus Floressein Anjiografi (FFA) ve Optik Koherens Tomografi (OCT) testlerinden yararlanmaktadır. Ancak, söz konusu testlerin doğru bir şekilde yorumlanabilmesi için alanında uzman hekimlerin değerlendirmesine ihtiyaç duyulmaktadır. Özellikle hasta sayısındaki artış, bu görüntülerin manuel olarak incelenmesini güçleştirmekte; buna ek olarak, kırsal bölgelerde yaşanan uzman hekim eksikliği, tanı süreçlerinde gecikmelere neden olabilmektedir. Bu çalışmada ise Retinal OCT-C8 veri seti kullanılarak 8 farklı retina hastalığı sınıflandırılmıştır. Sınıflandırma sürecinde standart You Only Look Once (YOLO) algoritmasının yanı sıra geliştirilmiş YOLO modelleri de kullanılmıştır. Geliştirilen YOLO modellerinde mimari derinlik katman sayısı artırılarak iyileştirilmiştir. YOLOv5 modelinde ek olarak K-Means ile kümelenmiş anchor boxlar eklenmiştir. Retinal OCT-C8 veri setinde yapılan test işleminde geliştirilmiş YOLOv8 modeli, %98.1 doğruluk, %97.6 kesinlik, %98.6 duyarlılık ve %98 F-Skor değerleri ile bu karşılaştırmadaki en iyi performans gösteren modeldir. Modelin genel performansını değerlendirmek amacıyla farklı veri setleri üzerinde de testler gerçekleştirilmiştir. OCTDL veri seti ile yapılan testlerde %88 doğruluk değeri ve %89, F-Skor değeri, OCTID veri seti ile yapılan testlerde ise %86 doğruluk değeri ve %90 F-Skor değeri elde edilmiştir. Ayrıca bu çalışmada, doktorların retina fotoğraflarını yükleyerek test edebileceği ve hasta bilgilerini düzenleyebileceği bir yazılım geliştirilmiştir. Elde edilen sonuçlar, geliştirilen modelin farklı veri kümeleri üzerinde yüksek doğrulukla çalıştığını ve retina hastalıklarının otomatik teşhisinde etkili bir yöntem sunduğunu ortaya koymaktadır. Ayrıca geliştirilen yazılım ile birlikte, tanı sürecinin kısa sürede tamamlanabileceğini de göstermektedir.
- 10 March 2026
- Published by Erciyes Universitesi (DergiPark) in Erciyes Üniversitesi Fen Bilimleri Enstitüsü Fen Bilimleri Dergisi
- Vol. 42 (1)
Göz Hastalıklarında Tanı Yöntemleri ve Derin Öğrenme Uygulamaları: Bir Literatür Çalışması
Göz hastalıklarının erken teşhisi, görme kaybını önlemek ve hastaların yaşam kalitesini korumak açısından kritik öneme sahiptir. Son yıllarda, biyomedikal görüntülemede derin öğrenme yöntemlerindeki hızlı ilerleme, oftalmoloji alanında önemli uygulamalar bulmuştur. Fundus fotoğrafları, optik koherens tomografi ve retina görüntüleri kullanılarak geliştirilen modeller, diyabetik retinopati, glokom, yaşa bağlı makula dejenerasyonu ve diğer retina temelli hastalıkların otomatik teşhisinde yüksek doğruluk oranlarına ulaşmıştır. Bu derleme, göz hastalıklarını tespit etmek için kullanılan mevcut derin öğrenme yaklaşımlarını, kullanılan veri kümelerini, mimari farklılıkları ve performans ölçütlerini incelemektedir. Evrişimsel sinir ağları, transfer öğrenme, hibrit modeller ve giderek artan şekilde görü dönüştürücü tabanlı yöntemler karşılaştırmalı olarak değerlendirilmektedir. Çalışma, alandaki son gelişmeleri özetlemeyi ve gelecekteki araştırmalara yön verecek öneriler sunmayı amaçlamaktadır.
- 10 March 2026
- Published by Erciyes Universitesi (DergiPark) in Erciyes Üniversitesi Fen Bilimleri Enstitüsü Fen Bilimleri Dergisi
- Vol. 42 (1)
Göz Hastalıklarında Tanı Yöntemleri ve Derin Öğrenme Uygulamaları: Bir Literatür Çalışması
Göz hastalıklarının erken teşhisi, görme kaybını önlemek ve hastaların yaşam kalitesini korumak açısından kritik öneme sahiptir. Son yıllarda, biyomedikal görüntülemede derin öğrenme yöntemlerindeki hızlı ilerleme, oftalmoloji alanında önemli uygulamalar bulmuştur. Fundus fotoğrafları, optik koherens tomografi ve retina görüntüleri kullanılarak geliştirilen modeller, diyabetik retinopati, glokom, yaşa bağlı makula dejenerasyonu ve diğer retina temelli hastalıkların otomatik teşhisinde yüksek doğruluk oranlarına ulaşmıştır. Bu derleme, göz hastalıklarını tespit etmek için kullanılan mevcut derin öğrenme yaklaşımlarını, kullanılan veri kümelerini, mimari farklılıkları ve performans ölçütlerini incelemektedir. Evrişimsel sinir ağları, transfer öğrenme, hibrit modeller ve giderek artan şekilde görü dönüştürücü tabanlı yöntemler karşılaştırmalı olarak değerlendirilmektedir. Çalışma, alandaki son gelişmeleri özetlemeyi ve gelecekteki araştırmalara yön verecek öneriler sunmayı amaçlamaktadır.
- Open Access
- 27 January 2026
- Published by Deu Muhendislik Fakultesi Fen ve Muhendislik in Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi
- Vol. 28 (82) , 148-156
Design and Production of a Prototype Conveyor Belt for Real-Time Surface Imaging and Data Set Creation
In this study, a conveyor belt-based prototype system was developed to create a data set by imaging surface deformations that may occur during the production process of solid aluminum conductors. The system operates using a high-resolution camera integrated into a vision inspection box with controlled lighting and an adjustable viewing angle, capable of capturing images at a resolution of 1280×1024 pixels at a rate of 60 frames per second. The images were collected on the conveyor belt moving at a constant speed, thereby ensuring continuity in image quality. With the developed prototype, a total of 1.600 images were collected and manually labeled as examples of three different defect types. The obtained dataset was organized in a way that will enable the detection and classification of surface defects such as pitting, embossing, and scalping using deep learning algorithms in subsequent stages. In this respect, the study provides the visual data infrastructure necessary for real-time artificial intelligence applications in aluminum conductor production. The prototype was designed using low-cost and easily obtainable hardware and offers a modular system that can be integrated into production lines. The main contribution of this study includes the development of a prototype system that operates under conditions suitable for real production environments. This system is equipped with automated image acquisition capabilities, ensuring consistent and scalable data collection. Additionally, this study provides a new dataset for detecting surface deformations in aluminum conductors. The developed infrastructure aims to form the basis for artificial intelligence-based solutions for industrial quality control systems.
- 27 January 2026
- Published by Deu Muhendislik Fakultesi Fen ve Muhendislik in Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi
- Vol. 28 (82) , 148-156
Open AccessDesign and Production of a Prototype Conveyor Belt for Real-Time Surface Imaging and Data Set Creation
In this study, a conveyor belt-based prototype system was developed to create a data set by imaging surface deformations that may occur during the production process of solid aluminum conductors. The system operates using a high-resolution camera integrated into a vision inspection box with controlled lighting and an adjustable viewing angle, capable of capturing images at a resolution of 1280×1024 pixels at a rate of 60 frames per second. The images were collected on the conveyor belt moving at a constant speed, thereby ensuring continuity in image quality. With the developed prototype, a total of 1.600 images were collected and manually labeled as examples of three different defect types. The obtained dataset was organized in a way that will enable the detection and classification of surface defects such as pitting, embossing, and scalping using deep learning algorithms in subsequent stages. In this respect, the study provides the visual data infrastructure necessary for real-time artificial intelligence applications in aluminum conductor production. The prototype was designed using low-cost and easily obtainable hardware and offers a modular system that can be integrated into production lines. The main contribution of this study includes the development of a prototype system that operates under conditions suitable for real production environments. This system is equipped with automated image acquisition capabilities, ensuring consistent and scalable data collection. Additionally, this study provides a new dataset for detecting surface deformations in aluminum conductors. The developed infrastructure aims to form the basis for artificial intelligence-based solutions for industrial quality control systems.
- Open Access
- 13 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 8484-8505
ExpAbsTanH: A Novel Activation Function for Image and Video-Based Human Activity Recognition
Activation functions are fundamental components of deep neural networks, providing the nonlinear transformations that enable complex representation learning and strongly influence model performance in deep learning–based artificial intelligence systems. However, despite their proven importance, commonly used activation functions often suffer from critical limitations such as vanishing or exploding gradients, non-differentiability at zero, and instability during training. These limitations become particularly relevant in image classification and Human Activity Recognition (HAR) tasks, where complex spatial or spatiotemporal dependencies, along with noisy data, can adversely affect training stability. To address these limitations, we propose a novel activation function, ExpAbsTanH, which integrates self-gating with exponential control and smooth nonlinearity. To assess its theoretical robustness and stability, initial experiments were conducted on the MNIST and CIFAR-10 datasets, demonstrating that ExpAbsTanH consistently delivers strong classification performance across diverse deep learning architectures. ExpAbsTanH was found to outperform ReLU and Swish in most cases, achieving performance comparable to that of Mish. When integrated into ResNet-200, ExpAbsTanH achieved a Top-1 accuracy of 90.08% on CIFAR-10, highlighting its stability and effectiveness in deep convolutional networks. Beyond static images, the robustness of ExpAbsTanH for HAR was evaluated on the UCF-50 and UCF-101 datasets. For temporally complex tasks, a hybrid model combining ResNet-50 with a single-layer Long Short-Term Memory (LSTM) network and an attention module was employed. On UCF-50, the ExpAbsTanH-based model consistently outperformed the ReLU-based model, achieving Top-1 and Top-5 accuracies of 95.69% and 98.74%, respectively. To further assess its generalization capability, experiments conducted on randomly selected subsets of the UCF-101 dataset demonstrated the highest Top-1 and Top-5 accuracies of 99.47% and 99.79%, respectively. Under noisy conditions, the ExpAbsTanH-based model achieved a Top-1 accuracy of 95.51% on the UCF-50 dataset, consistently surpassing the ReLU-based model. On the UCF-101 dataset, this integration maintained a Top-1 accuracy of 98.63%, demonstrating its robustness against input distortions. Gradient-weighted Class Activation Mapping (Grad-CAM) analysis further revealed that ExpAbsTanH focused on semantically meaningful and spatially coherent features, whereas ReLU occasionally emphasized irrelevant regions. ExpAbsTanH has become a robust and effective alternative activation function, improving the reliability of real-world deep learning systems by delivering high performance across image and video recognition tasks.
- 13 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 8484-8505
Open AccessExpAbsTanH: A Novel Activation Function for Image and Video-Based Human Activity Recognition
Activation functions are fundamental components of deep neural networks, providing the nonlinear transformations that enable complex representation learning and strongly influence model performance in deep learning–based artificial intelligence systems. However, despite their proven importance, commonly used activation functions often suffer from critical limitations such as vanishing or exploding gradients, non-differentiability at zero, and instability during training. These limitations become particularly relevant in image classification and Human Activity Recognition (HAR) tasks, where complex spatial or spatiotemporal dependencies, along with noisy data, can adversely affect training stability. To address these limitations, we propose a novel activation function, ExpAbsTanH, which integrates self-gating with exponential control and smooth nonlinearity. To assess its theoretical robustness and stability, initial experiments were conducted on the MNIST and CIFAR-10 datasets, demonstrating that ExpAbsTanH consistently delivers strong classification performance across diverse deep learning architectures. ExpAbsTanH was found to outperform ReLU and Swish in most cases, achieving performance comparable to that of Mish. When integrated into ResNet-200, ExpAbsTanH achieved a Top-1 accuracy of 90.08% on CIFAR-10, highlighting its stability and effectiveness in deep convolutional networks. Beyond static images, the robustness of ExpAbsTanH for HAR was evaluated on the UCF-50 and UCF-101 datasets. For temporally complex tasks, a hybrid model combining ResNet-50 with a single-layer Long Short-Term Memory (LSTM) network and an attention module was employed. On UCF-50, the ExpAbsTanH-based model consistently outperformed the ReLU-based model, achieving Top-1 and Top-5 accuracies of 95.69% and 98.74%, respectively. To further assess its generalization capability, experiments conducted on randomly selected subsets of the UCF-101 dataset demonstrated the highest Top-1 and Top-5 accuracies of 99.47% and 99.79%, respectively. Under noisy conditions, the ExpAbsTanH-based model achieved a Top-1 accuracy of 95.51% on the UCF-50 dataset, consistently surpassing the ReLU-based model. On the UCF-101 dataset, this integration maintained a Top-1 accuracy of 98.63%, demonstrating its robustness against input distortions. Gradient-weighted Class Activation Mapping (Grad-CAM) analysis further revealed that ExpAbsTanH focused on semantically meaningful and spatially coherent features, whereas ReLU occasionally emphasized irrelevant regions. ExpAbsTanH has become a robust and effective alternative activation function, improving the reliability of real-world deep learning systems by delivering high performance across image and video recognition tasks.
- 15 July 2025
- Published by Omer Halisdemir Universitesi in Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 14 (3) , 981-989
Robotik süreç otomasyonu ile üniversite süreçlerinin dönüşümü: Erciyes Üniversitesi örneği
Robotik Süreç Otomasyonu (Robotic Process Automation, RPA), rutin görevlerin yürütme süresini kısaltarak çalışanların daha yaratıcı görevlere odaklanmasını sağladığı için dijital dönüşüm süreçlerinde önemli bir teknoloji haline gelmiştir. 2012’den beri kullanılan RSO, iş süreçlerinin yeniden yapılandırılmasında yaygın olarak tercih edilmektedir. Bu çalışmada, Erciyes Üniversitesi bünyesindeki Öğrenci İşleri Daire Başkanlığı(ÖİDB) ve Bilgi İşlem Daire Başkanlığı(BİDB) gibi birimlerin öğrenci kayıtları, not girişleri ve burs başvuruları gibi zaman alan tekrarlayan işlemleri incelenmiş ve RSO ile otomasyona uygun süreçler belirlenmiştir. Manuel işlemlerden kaynaklanan iş yükünün azaltılması ve süreçlerin verimliliğinin artırılması hedeflenmiştir. Çalışma, RSO’nun üniversiteye sağladığı faydaları değerlendirerek kamu kurumları ve üniversitelere yol gösterici olmayı amaçlamaktadır.
- 15 July 2025
- Published by Omer Halisdemir Universitesi in Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi
- Vol. 14 (3) , 981-989
Robotik süreç otomasyonu ile üniversite süreçlerinin dönüşümü: Erciyes Üniversitesi örneği
Robotik Süreç Otomasyonu (Robotic Process Automation, RPA), rutin görevlerin yürütme süresini kısaltarak çalışanların daha yaratıcı görevlere odaklanmasını sağladığı için dijital dönüşüm süreçlerinde önemli bir teknoloji haline gelmiştir. 2012’den beri kullanılan RSO, iş süreçlerinin yeniden yapılandırılmasında yaygın olarak tercih edilmektedir. Bu çalışmada, Erciyes Üniversitesi bünyesindeki Öğrenci İşleri Daire Başkanlığı(ÖİDB) ve Bilgi İşlem Daire Başkanlığı(BİDB) gibi birimlerin öğrenci kayıtları, not girişleri ve burs başvuruları gibi zaman alan tekrarlayan işlemleri incelenmiş ve RSO ile otomasyona uygun süreçler belirlenmiştir. Manuel işlemlerden kaynaklanan iş yükünün azaltılması ve süreçlerin verimliliğinin artırılması hedeflenmiştir. Çalışma, RSO’nun üniversiteye sağladığı faydaları değerlendirerek kamu kurumları ve üniversitelere yol gösterici olmayı amaçlamaktadır.
- Open Access
- 27 June 2025
- Published by Orclever Science and Research Group in The European Journal of Research and Development
- Vol. 5 (1) , 52-67
A Review of Deep Learning Approaches with CMR Images in the Diagnosis of Cardiovascular Diseases
Cardiovascular disease (CVD) is one of the leading causes of death worldwide, which has led to the recent intensification of Deep Learning (DL) studies in the field of cardiology. Patients usually experience symptoms such as rapid fatigue, edema below the knee and ankle, chest pain, shortness of breath, and palpitations. The most common types of CVD include coronary artery disease, arrhythmias, congenital heart defects, cardiomyopathy, valvular heart failure, and angina. Electrocardiography (ECG), blood tests, physical examination, and medical imaging are the most effective tools for diagnosing diseases. In recent years, cardiac magnetic resonance imaging (CMRI) has been increasingly used for the diagnosis, follow-up, treatment planning, and prognosis of CVDs. However, the large number of slices and low contrast of CMRI data make diagnosing CVD difficult. Deep learning techniques are being applied to diagnose CVD with CMRI data to overcome these difficulties, and intensive research continues to be conducted in this field. It is important to keep abreast of developments so that these studies can significantly impact clinical applications. This review aims to be a stepping stone for researchers in this process by comprehensively reviewing studies on CVD detection using DL methods on CMRI images.
- 27 June 2025
- Published by Orclever Science and Research Group in The European Journal of Research and Development
- Vol. 5 (1) , 52-67
Open AccessA Review of Deep Learning Approaches with CMR Images in the Diagnosis of Cardiovascular Diseases
Cardiovascular disease (CVD) is one of the leading causes of death worldwide, which has led to the recent intensification of Deep Learning (DL) studies in the field of cardiology. Patients usually experience symptoms such as rapid fatigue, edema below the knee and ankle, chest pain, shortness of breath, and palpitations. The most common types of CVD include coronary artery disease, arrhythmias, congenital heart defects, cardiomyopathy, valvular heart failure, and angina. Electrocardiography (ECG), blood tests, physical examination, and medical imaging are the most effective tools for diagnosing diseases. In recent years, cardiac magnetic resonance imaging (CMRI) has been increasingly used for the diagnosis, follow-up, treatment planning, and prognosis of CVDs. However, the large number of slices and low contrast of CMRI data make diagnosing CVD difficult. Deep learning techniques are being applied to diagnose CVD with CMRI data to overcome these difficulties, and intensive research continues to be conducted in this field. It is important to keep abreast of developments so that these studies can significantly impact clinical applications. This review aims to be a stepping stone for researchers in this process by comprehensively reviewing studies on CVD detection using DL methods on CMRI images.
- Open Access
- 22 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 71822-71839
BorB: A Novel Image Segmentation Technique for Improving Plant Disease Classification With Deep Learning Models
Disease detection from leaf images has been among the popular studies in recent years. Classifying leaf diseases using computational methods provides great convenience for farming. In the studies carried out in this field, systems that work with high accuracy and are least affected by environmental factors that can be used in agricultural lands come to the fore. This study investigates the application of deep learning architectures for accurate and efficient plant disease detection within the context of the ongoing digital transformation of the agricultural sector. Recognizing the critical role of AI in modernizing agriculture, this research focuses on enhancing the accuracy of the classification of plant diseases. To facilitate this research, a novel dataset, "EruCauliflowerDB", was meticulously curated, comprising high-resolution images of cauliflower plants infected with Alternaria Leaf Spot and Black Rot. The obtained EruCauliflower dataset contains 114 images from the Alternaria Leaf Spot disease class and 99 images from the Black Rot disease class. A novel integrated classification system was developed, encompassing three key stages. First, a novel segmentation method, "BorB," was introduced to effectively isolate diseased leaf regions. This segmentation method enables us to extract features of leaf images in Lab and RGB formats. Combining the features obtained from the two image formats with the OR logical operation separates the leaf region from the background. Second, data augmentation techniques, including geometric transformations, were applied to the segmented images to enhance data diversity and improve model robustness. Finally, four state-of-the-art deep learning models—VGG16, ResNet50, EfficientNetB3, and MobileNetV3 Large—were employed for disease classification. The proposed integrated system demonstrated exceptional performance, achieving 100% classification accuracy on the EruCauliflowerDB dataset across all four models. To assess the system’s robustness, further evaluations were conducted on the independent MangoLeafBD dataset, yielding consistent results with 100% classification accuracy. The proposed Integrated Classifier method was applied by selecting 15 classes from the PlantVillage, another multi-class dataset. As a result of the experiments, PlantVillage plant leaf images were classified with 99.78% accuracy. Experimental results show that the proposed method can be effectively utilized in real-world agricultural settings to assist farmers in early disease detection, thereby reducing crop losses and improving yield quality.
- 22 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 71822-71839
Open AccessBorB: A Novel Image Segmentation Technique for Improving Plant Disease Classification With Deep Learning Models
Disease detection from leaf images has been among the popular studies in recent years. Classifying leaf diseases using computational methods provides great convenience for farming. In the studies carried out in this field, systems that work with high accuracy and are least affected by environmental factors that can be used in agricultural lands come to the fore. This study investigates the application of deep learning architectures for accurate and efficient plant disease detection within the context of the ongoing digital transformation of the agricultural sector. Recognizing the critical role of AI in modernizing agriculture, this research focuses on enhancing the accuracy of the classification of plant diseases. To facilitate this research, a novel dataset, "EruCauliflowerDB", was meticulously curated, comprising high-resolution images of cauliflower plants infected with Alternaria Leaf Spot and Black Rot. The obtained EruCauliflower dataset contains 114 images from the Alternaria Leaf Spot disease class and 99 images from the Black Rot disease class. A novel integrated classification system was developed, encompassing three key stages. First, a novel segmentation method, "BorB," was introduced to effectively isolate diseased leaf regions. This segmentation method enables us to extract features of leaf images in Lab and RGB formats. Combining the features obtained from the two image formats with the OR logical operation separates the leaf region from the background. Second, data augmentation techniques, including geometric transformations, were applied to the segmented images to enhance data diversity and improve model robustness. Finally, four state-of-the-art deep learning models—VGG16, ResNet50, EfficientNetB3, and MobileNetV3 Large—were employed for disease classification. The proposed integrated system demonstrated exceptional performance, achieving 100% classification accuracy on the EruCauliflowerDB dataset across all four models. To assess the system’s robustness, further evaluations were conducted on the independent MangoLeafBD dataset, yielding consistent results with 100% classification accuracy. The proposed Integrated Classifier method was applied by selecting 15 classes from the PlantVillage, another multi-class dataset. As a result of the experiments, PlantVillage plant leaf images were classified with 99.78% accuracy. Experimental results show that the proposed method can be effectively utilized in real-world agricultural settings to assist farmers in early disease detection, thereby reducing crop losses and improving yield quality.
- Open Access
- 1 March 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (5) , 2648
Baby Cry Classification Using Structure-Tuned Artificial Neural Networks with Data Augmentation and MFCC Features
Babies express their needs, such as hunger, discomfort, or sleeplessness, by crying. However, understanding these cries correctly can be challenging for parents. This can delay the baby’s needs, increase parents’ stress levels, and negatively affect the baby’s development. In this paper, an integrated system for the classification of baby sounds is proposed. The proposed method includes data augmentation, feature extraction, hyperparameter tuning, and model training steps. In the first step, various data augmentation techniques were applied to increase the training data’s diversity and strengthen the model’s generalization capacity. The MFCC (Mel-Frequency Cepstral Coefficients) method was used in the second step to extract meaningful and distinctive features from the sound data. MFCC represents sound signals based on the frequencies the human ear perceives and provides a strong basis for classification. The obtained features were classified with an artificial neural network (ANN) model with optimized hyperparameters. The hyperparameter optimization of the model was performed using the grid search algorithm, and the most appropriate parameters were determined. The training, validation, and test data sets were separated at 75%, 10%, and 15% ratios, respectively. The model’s performance was tested on mixed sounds. The test results were analyzed, and the proposed method showed the highest performance, with a 90% accuracy rate. In the comparison study with an artificial neural network (ANN) on the Donate a Cry data set, the F1 score was reported as 46.99% and the test accuracy as 85.93%. In this paper, additional techniques such as data augmentation, hyperparameter tuning, and MFCC feature extraction allowed the model accuracy to reach 90%. The proposed method offers an effective solution for classifying baby sounds and brings a new approach to this field.
- 1 March 2025
- Published by MDPI AG in Applied Sciences
- Vol. 15 (5) , 2648
Open AccessBaby Cry Classification Using Structure-Tuned Artificial Neural Networks with Data Augmentation and MFCC Features
Babies express their needs, such as hunger, discomfort, or sleeplessness, by crying. However, understanding these cries correctly can be challenging for parents. This can delay the baby’s needs, increase parents’ stress levels, and negatively affect the baby’s development. In this paper, an integrated system for the classification of baby sounds is proposed. The proposed method includes data augmentation, feature extraction, hyperparameter tuning, and model training steps. In the first step, various data augmentation techniques were applied to increase the training data’s diversity and strengthen the model’s generalization capacity. The MFCC (Mel-Frequency Cepstral Coefficients) method was used in the second step to extract meaningful and distinctive features from the sound data. MFCC represents sound signals based on the frequencies the human ear perceives and provides a strong basis for classification. The obtained features were classified with an artificial neural network (ANN) model with optimized hyperparameters. The hyperparameter optimization of the model was performed using the grid search algorithm, and the most appropriate parameters were determined. The training, validation, and test data sets were separated at 75%, 10%, and 15% ratios, respectively. The model’s performance was tested on mixed sounds. The test results were analyzed, and the proposed method showed the highest performance, with a 90% accuracy rate. In the comparison study with an artificial neural network (ANN) on the Donate a Cry data set, the F1 score was reported as 46.99% and the test accuracy as 85.93%. In this paper, additional techniques such as data augmentation, hyperparameter tuning, and MFCC feature extraction allowed the model accuracy to reach 90%. The proposed method offers an effective solution for classifying baby sounds and brings a new approach to this field.
- Open Access
- 1 January 2024
- Published by Kare Publishing in Journal of Clinical Practice and Research
- Vol. 46 (6) , 527-536
Applications of Deep Learning Techniques in Healthcare Systems: A Review
Artificial intelligence (AI) is the ability of machines to carry out tasks by imitating human intelligence. In recent years, AI methods have begun to be applied in many different areas, with healthcare being one of the most prominent. Diagnosis, treatment, patient care, new drug production, and preventive care can be listed as some of the applications of AI in healthcare. In this review, deep learning methods, which are a sub-branch of AI, are mentioned. Deep learning methods frequently used in the literature are convolutional neural networks (CNNs), stacked autoencoders (SAEs), and recurrent neural networks (RNNs). These deep learning methods include CNNs for image recognition and classification, SAEs for unsupervised feature learning and dimensionality reduction, and RNNs for analyzing sequential data like time-series. However, it should be noted that these methods can also be applied to other application areas. This paper presents studies in the literature on medical image analysis, drug discovery and development, and remote patient monitoring in which these deep learning methods are used.
- 1 January 2024
- Published by Kare Publishing in Journal of Clinical Practice and Research
- Vol. 46 (6) , 527-536
Open AccessApplications of Deep Learning Techniques in Healthcare Systems: A Review
Artificial intelligence (AI) is the ability of machines to carry out tasks by imitating human intelligence. In recent years, AI methods have begun to be applied in many different areas, with healthcare being one of the most prominent. Diagnosis, treatment, patient care, new drug production, and preventive care can be listed as some of the applications of AI in healthcare. In this review, deep learning methods, which are a sub-branch of AI, are mentioned. Deep learning methods frequently used in the literature are convolutional neural networks (CNNs), stacked autoencoders (SAEs), and recurrent neural networks (RNNs). These deep learning methods include CNNs for image recognition and classification, SAEs for unsupervised feature learning and dimensionality reduction, and RNNs for analyzing sequential data like time-series. However, it should be noted that these methods can also be applied to other application areas. This paper presents studies in the literature on medical image analysis, drug discovery and development, and remote patient monitoring in which these deep learning methods are used.
- 17 April 2023
- Published by Omer Halisdemir Universitesi in Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi
Fetüs Beyin Dokularının 3B U-Net ile Segmentasyonu ve Gestasyonel Yaşın Segmentasyon Performansına Etkisi
Gebelik esnasında fetüs beyninde, çevresel veya genetik etmenlerden kaynaklanan bozuklukların ilerleyen yaşlarda otizm, hiperaktivite ve bipolar bozukluklar olarak ortaya çıktığı düşünülmektedir. Fetüs beyin dokularının yapısal analizi, dokuların büyüklükleri ve şekilleri hakkında bilgi sağlayarak bu hastalıkların etiyolojisinin araştırılmasına yardımcı olmaktadır. MR (manyetik rezonans) ve ultrason fetüs beyin dokusu analizi için en sık kullanılan iki görüntüleme tekniğidir. Özellikle MR görüntülemede, fetüs beyin dokuları net bir şekilde görülebilmesine rağmen bu görüntülerin yapısal analizi zaman alıcı bir iştir. Hızla gelişen ve değişen fetüs beyin dokuları, haftalık veya aylık periyotlarla, elle veya yarı-otomatik şekilde gerçekleştirilen MR analizinin yapılmasını zorlaştırmaktadır. Bu araştırmada, 3B U-Net ile MR görüntüleri üzerinde yedi farklı fetüs beyin dokusunun tam otomatik segmentasyonu gerçekleştirilmiştir. FeTA2021 veri seti, erken, orta ve geç gestasyonel yaş gruplarına bölünerek her bir yaş grubu için farklı 3B U-Net modelleri eğitilmiş ve segmentasyon performansı analiz edilmiştir. Gestasyonel yaş gruplarına göre sırasıyla ortalama 0,83, 0,91 ve 0,92 Dice skoruna ulaşılmıştır.
- 17 April 2023
- Published by Omer Halisdemir Universitesi in Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi
Fetüs Beyin Dokularının 3B U-Net ile Segmentasyonu ve Gestasyonel Yaşın Segmentasyon Performansına Etkisi
Gebelik esnasında fetüs beyninde, çevresel veya genetik etmenlerden kaynaklanan bozuklukların ilerleyen yaşlarda otizm, hiperaktivite ve bipolar bozukluklar olarak ortaya çıktığı düşünülmektedir. Fetüs beyin dokularının yapısal analizi, dokuların büyüklükleri ve şekilleri hakkında bilgi sağlayarak bu hastalıkların etiyolojisinin araştırılmasına yardımcı olmaktadır. MR (manyetik rezonans) ve ultrason fetüs beyin dokusu analizi için en sık kullanılan iki görüntüleme tekniğidir. Özellikle MR görüntülemede, fetüs beyin dokuları net bir şekilde görülebilmesine rağmen bu görüntülerin yapısal analizi zaman alıcı bir iştir. Hızla gelişen ve değişen fetüs beyin dokuları, haftalık veya aylık periyotlarla, elle veya yarı-otomatik şekilde gerçekleştirilen MR analizinin yapılmasını zorlaştırmaktadır. Bu araştırmada, 3B U-Net ile MR görüntüleri üzerinde yedi farklı fetüs beyin dokusunun tam otomatik segmentasyonu gerçekleştirilmiştir. FeTA2021 veri seti, erken, orta ve geç gestasyonel yaş gruplarına bölünerek her bir yaş grubu için farklı 3B U-Net modelleri eğitilmiş ve segmentasyon performansı analiz edilmiştir. Gestasyonel yaş gruplarına göre sırasıyla ortalama 0,83, 0,91 ve 0,92 Dice skoruna ulaşılmıştır.