Sürdürülebilir Kalkınma Amaçları
Yazarlar
- Dogan, Ramazan Ozgur(6)
- Dogan, Hulya(5)
- Catal Reis, Hatice(4)
- Reis, Hatice Catal(4)
- Turk, Veysel(4)
- Bozkurt, Ferhat(3)
- Celik, Kemal(3)
- İncir, Ramazan(3)
- Bingol, Ozkan(2)
- Gökgöz, Baki(2)
- Gül, Fatih(2)
- Ture, Hayati(2)
- Al-Taib, Asmaa Abduh(1)
- Almasani, Heba(1)
- Alparslan Kiliç(1)
- Aslan, Muhammed Emin(1)
- Ay, Ilyas(1)
- Aydin, Tolga(1)
- Aydın, Tolga(1)
- Bayrak, Coskun(1)
- Başağa, Hasan Basri(1)
- Başoğlu, Mustafa Engin(1)
- Bingöl, Ozkan(1)
- Bulut, Sinan(1)
- CELIK, Ayse(1)
- CELIK, Fatih(1)
- CELIK, Kemal(1)
- Cam, Handan(1)
- Celik, Fatih(1)
- Cetin, Gokhan(1)
- Demir, Yasin(1)
- Demir, Zeynep(1)
- Demirel, Ugur(1)
- Dogu, Fatma Tuana(1)
- Ekinci, Murat(1)
- Ekmekci, Dursun(1)
- Esme, Engin(1)
- Gedikli, Eyup(1)
- Genemo, Musa Dima(1)
- Günen, Mehmet Akif(1)
Yayınlar
- Yüksek Lisans TeziYalnız künye2026Akıllı telefon görüntüleri kullanılarak yapılardaki kırıkların otomatik tespiti
Alparslan Kiliç
Yapılardaki çatlakların/kırıkların erken ve doğru tespiti, yapı sağlığı ve bakım süreçlerinin planlanması açısından önem taşımaktadır. Bu çalışmada, akıllı telefon görüntüleri kullanılarak duvar ve köprü yüzeylerindeki çatlakların derin öğrenme yöntemleriyle otomatik olarak tespit edilmesi amaçlanmıştır. Duvar ve köprü veri setleri “çatlak” ve “çatlaksız” sınıflardan oluşmaktadır. Yerel veri seti ile SDNET2018 üzerinde evrişimli sinir ağı, dönüştürücü ve hibrit yapılardan oluşan on sekiz mimari; doğruluk, pozitif kestirim değeri, duyarlılık, özgüllük, F1 skoru, AUC ve görüntü başına çıkarım süresi bakımından karşılaştırılmıştır. Yerel deneylerde önerilen model, duvar veri setinde %98,50 ve köprü veri setinde %100,00 doğruluk sağlamıştır. Buna karşılık sıfır örnekli dış veri seti testlerinde doğruluk %52,43–%73,53 aralığına düşmüştür. Ham ortak eğitim sonrasında SDNET2018 duvar, köprü tabliyesi ve yol kaplama testlerinde sırasıyla %86,05, %81,35 ve %87,21 doğruluk elde edilmiştir; dengeli ortak eğitimde bu değerler %82,84, %80,86 ve %86,57 olarak gerçekleşmiştir. Bulgular, veri kaynağı farklılıklarının model genellemesi üzerinde belirleyici olduğunu ve çoklu alan ortak eğitimin dış veri seti başarısını artırabildiğini göstermektedir. Sonuç olarak, akıllı telefon görüntülerine dayalı çatlak/kırık tespiti afetlerde ve yapı sağlığı yönetiminde uzman değerlendirmesini destekleyen hızlı ve düşük maliyetli bir yaklaşım olarak sunulabilir.
- MakaleYalnız künye2026Optic disc segmentation: Improved U-Net model and hyperparameter optimization with harmony search algorithm
İncir, Ramazan; Bozkurt, Ferhat
Vision loss negatively impacts individuals quality of life, creating a societal issue. As stated by the World Health Organization, vision loss is significant concern among health problems. Glaucoma and Diabetic Retinopathy are among the primary causes of vision impairment and loss. Early diagnosis of these diseases is crucial for guiding appropriate treatment. Optical disc segmentation plays a key role in diagnosing these diseases and helps in effectively detecting them through the analysis of retinal images. This study proposes method based on the automatic processing of fundus images to contribute to the early diagnosis of eye diseases. Improved U-Net architecture developed for optical disc segmentation is used. To enhance the models performance, the improved U-Net architecture has been strengthened with residual and custom convolution blocks. Genetic Algorithm, Particle Swarm Optimization Algorithm, Ant Colony Optimization and Harmony Search Algorithm were used for hyperparameter optimization. Comparisons showed that the Harmony Search Algorithm achieved better results, improving overall model performance and providing more accurate segmentation outcomes. The models performance has been tested on the IDRiD and DRISHTI-GS datasets. The improved architecture outperformed existing methods in the literature. Significant improvements were observed in key segmentation metrics such as IoU and Dice coefficient. In experimental studies using the IDRiD and DRISHTI-GS datasets, the proposed model achieved IoU scores of 93.65% and 95.20% and Dice scores of 96.72% and 97.54% respectively. The model was able to accurately define optical disc boundaries even in low contrast and complex images. The obtained success demonstrates the models potential as valuable tool for the early diagnosis of vision loss causing diseases.
- MakaleYalnız künye2026Electromechanical characterization and deep learning-based force prediction in 3D-printed PLA core–shell composites
Ekmekci, Dursun; Kaplan, Emrah
In this study, composite samples were produced using fused deposition modeling to contain an outer layer of red insulating polylactic acid and an inner core of black conductive polylactic acid. The mechanical properties of the samples were determined using three-point bending experiments, and their electro-mechanical impedance responses during loading were monitored using impedance experiments. To examine the effects of physical aging and crystallization, the composites were heat-treated at 50 °C for 2, 4, and 6 hours. The insulating polylactic acid exhibited brittle fracture with a sharp post-peak force drop, whereas the conductive polylactic acid showed a ductile fracture mode accompanied by substantial energy dissipation. The physically aged composites were stiffer and less ductile, which was attributed to the presence of partially crystallized and partially relaxed molecules, especially around the glass transition region. In support of the observed properties, scanning electron microscopy images confirmed that the fractured layers were smoother compared to the ductile conductive layers, which had surface roughness patterns resembling fibrillated or lamellar structures. Additionally, the impedance signals obtained during the bending experiments were linked with the corresponding mechanical properties using an advanced deep learning model that could model the electric signals to accurately provide the corresponding real-time force values.
- MakaleYalnız künye2026CropHealthNet: Potato Disease Detection Using Depthwise Separable Convolutional Neural Networks
Catal Reis, Hatice
Plant diseases are one of the primary factors that threaten food security, crop yields, harvesting, and sustainable production in the agricultural sector. In this context, early disease diagnosis is essential for enhancing agricultural productivity and reducing crop losses. In modern agricultural practices, machine learning-based approaches offer effective and innovative solutions for detecting and classifying plant diseases. These modern and innovative methods generally reach high accuracy rates with large-scale image datasets. In particular, deep learning algorithms have achieved significant success in modeling disease’s spread dynamics and detecting early disease symptoms. In this study, a convolutional neural network (CNN)-based, lightly parallel deep learning model called CropHealthNet is proposed for classifying potato leaf diseases. The model is designed with special depthwise separable convolution-based convolution layers to reduce the computational cost of CNN. CropHealthNet is a low-parameter model suitable for real-time operation, capable of recognizing disease-specific patterns with high accuracy. CropHealthNet aims to achieve deep and comprehensive feature extraction through its parallel structure and varying channel numbers. The features from each parallel structure are combined at multiple levels, increasing the model’s ability to detect abstract and deep patterns. Moreover, the proposed model has extremely low computational complexity with only 0.47 million parameters and 0.59 GFLOPs, thus offering high efficiency compared to existing approaches. The effectiveness of the proposed model was evaluated using the PlantVillage Potato Dataset. Since the PlantVillage Potato Dataset has an unbalanced data distribution, the number of data for each class has been equalized using the Synthetic Minority Over-sampling Technique (SMOTE) to evaluate the model’s performance against data imbalance. The proposed model succeeded in achieving 99.30% and 99.50% accuracy rates in the tests conducted with unbalanced and balanced datasets, respectively. Experimental findings show that the proposed model generalizes well despite data imbalance and correctly classify diseases on potato leaves. Therefore, the model can demonstrate strong performance in agricultural practices and can detect plant diseases under real-world conditions.
- MakaleYalnız künye2026A Dense and Sparse Attention in Vision Transformer for Remote Sensing Image Classification
Celik, Kemal
Satellite imagery plays a crucial role in various applications, such as environmental monitoring, urban planning, and disaster response. Previously, Convolutional neural networks have played a crucial role in improving remote sensing scene classification by effectively capturing spatial features. However, they often struggle with long-range dependencies and complex spectral patterns in satellite images. Recently, transformer-based architectures, initially designed for natural language processing, have been adopted in computer vision to leverage self-attention mechanisms for capturing long-range dependencies. To effectively extract the long-range dependencies and complex patterns, this paper proposes a novel method combining dense and sparse attention mechanisms to enhance remote sensing scene classification in different scales. The dense attention mechanisms capture comprehensive connections within satellite images, followed by sparse attention mechanisms applied at both tile and patch scale levels. This multi-scale attention strategy allows the model to efficiently focus on important spatial and spectral features while managing computational complexity. Extensive experimental evaluations on three benchmark datasets, such as Merced, AID, and Optimal-31, demonstrate significant improvements in classification accuracy, achieving 98.9, 97.5, and 97.8, respectively. Experimental results show that the proposed framework outperforms CNN and Transformer baselines, enhancing the accuracy and robustness of remote sensing scene classification for geospatial analysis.
- MakaleYalnız künye2026Multi level throttled attention for accurate and efficient classification of retinal diseases in OCT images
Celik, Kemal; Celik, Fatih
The classification of retinal diseases using Optical Coherence Tomography images has gained significant attention in recent years. However, due to the heterogeneity of retinal disorders, achieving high classification accuracy remains challenging. While deep learning has improved the identification of various ophthalmic conditions, accurate and early diagnosis is still critical for effective clinical decision-making. Additionally, reducing computational complexity is essential for building scalable diagnostic systems. To address these challenges, we propose a novel Multi-level Throttled Attention (MLTA) model that first applies attention to extract important features, followed by throttling to filter out redundant or less useful features. This process is applied at multiple stages, operating on both Convolutional Neural Network layers and Transformer architectures, and then fuses them finally. This approach optimizes feature selection by preserving the most informative retinal features while discarding redundant ones from multi-level feature maps. As a result, the model enhances classification performance while reducing computational overhead. We evaluate our proposed model on two benchmark OCT datasets: OCT2017 and OCTID, achieving accuracy scores of 99.59 ± 0.04% and 99.71 ± 0.03%, respectively. Our method outperforms existing CNN-based and Vision Transformer-based models, demonstrating superior accuracy and efficiency through this hierarchical attention.
- Yüksek Lisans TeziAçık Erişim2025Evrişimli sinir ağları tabanlı süper çözünürlük yöntemleri ile avuç izi görüntü kalitesinin iyileştirilmesi
Hüseyin Furkan Macan
Bu tez çalışması, avuç izi tanıma sistemlerinde kullanılan biyometrik görüntülerin kalitesini iyileştirmek amacıyla, derin öğrenme tabanlı süper-çözünürlük yöntemlerinin performansını incelemektedir. Literatürde yaygın olarak kullanılan GAN, DCGAN, BGAN, BEGAN ve özellikle çekişmeli otokodlayıcı (AAE) mimarileri değerlendirilmiş; elde edilen süper-çözünürlüklü çıktılar hem PSNR gibi referanslı hem de BRISQUE gibi referanssız metriklerle analiz edilmiştir. Yapılan deneylerde, AAE modeli en başarılı sonuçları verdiğinden, geliştirme süreci bu yapı üzerine odaklanmıştır. Modelin kod çözücü kısmı konvolüsyonel katmanlarla yeniden tasarlanarak 2 kat çözünürlük artırımı sağlayacak şekilde optimize edilmiş, IIT Delhi avuç içi veri seti ile eğitilmiştir. Görüntü kalitesi iyileştirme sürecinin yalnızca görsel kaliteyi değil, aynı zamanda biyometrik tanıma doğruluğunu da artırdığı yapılan testlerle ortaya konmuştur.
- MakaleAçık Erişim2025Advanced brain tumor analysis: a novel strategy for segmentation and classification using modern computational methods
Reis, Hatice Catal; Turk, Veysel
Despite their low incidence, brain tumors are one of the most invasive cancer types, constituting a significant burden of death and disease in all age groups. Early and accurate diagnosis of brain tumors plays a vital role in reducing mortality rates. The heterogeneous nature of brain tumors and the diversity of tumor lesions may make it difficult for radiologists to make the right decision in the manual diagnosis process. This study proposes the use of machine learning methods for the classification of brain tumors (pituitary, meningioma, and glioma) and the use of metaheuristic algorithms graph theory, and random walker algorithms in the segmentation of brain tumors. The classification performed with the proposed method obtained an overall accuracy rate of 98.33%. In addition, the classification accuracy of 99.50%, 99.50%, 98.67%, and 99.00% was achieved for no tumor, pituitary, meningioma, and glioma, respectively. Experiments in the segmentation process show that metaheuristic algorithms and max-flow graph cut approach produce successful results.
- BildiriYalnız künye2025Improving Skin Lesion Classification with Pretrained Deep Learning Models
Almasani, Heba; Ture, Hayati
Skin cancer is the most common type of cancer. Cancer can be defined as a disease characterized by the uncontrolled division and spread of cells. As with other types of cancer, early diagnosis and treatment play a crucial role in combating skin cancer. Since diagnosis is made through visual examination, factors such as the level of expertise and image artifacts negatively affect diagnostic accuracy and slow down the process.Computer-aided applications used to analyze such images provide faster and more accurate results. Additionally, these methods eliminate factors that reduce diagnostic accuracy during the classification and segmentation of skin lesions.This study classifies skin cancer to benign or malignant using deep learning and machine learning methods. In particular, Convolutional Neural Networks (CNNs) have shown promising results in medical image classification. The study evaluates six pre-trained CNN models - VGG16, EfficientNetB7, EfficientNetV2L, DenseNet, ConvNeXtBase and MobileNet - for binary skin lesion classification using the ISIC 2019 dataset. The study first preprocesses and balances the data to enhance performance. The results indicate that the ConvNeXtBase model achieved the highest success with accuracy rate of 84.63% after balancing the data that we have.
- BildiriYalnız künye2025Hybrid CNN-LSTM Framework for Automated Detection of Pediatric Heart Murmurs
Al-Taib, Asmaa Abduh; Ture, Hayati; Aslan, Muhammed Emin
This study presents a comprehensive deep learning framework for automated heart murmur detection from phonocardiogram (PCG) signals to address the subjective limitations of cardiac auscultation. Using the CirCor DigiScope dataset of 5,272 pediatric heart sound recordings from 1,568 subjects, we systematically evaluated multiple approaches including traditional machine learning models (Logistic Regression, SVM, KNN, Decision Tree, AdaBoost), basic deep learning architectures (standalone CNN and LSTM), and advanced transfer learning models (DenseNet121, EfficientNet B0/B1/B2, ResNet 50/101, MobileNet V2/V3, GhostNet, AttentionCNN). Our hybrid CNN-LSTM architecture combines mel-spectrogram spatial feature extraction with temporal sequence modeling using 128 traditional audio features and comprehensive preprocessing with data augmentation. The CNN-LSTM model achieved superior performance with 84.6% accuracy, 0.84 F1-score, 0.748 precision, and 0.800 recall, significantly outperforming traditional methods (best SVM: 0.716 F1-score) and advanced architectures (best DenseNet121: 0.72 F1-score), demonstrating robust classification with 0.879 AUC. Results indicate that hybrid deep learning architectures effectively automate cardiac auscultation, providing objective diagnostic tools with potential for improved healthcare accessibility and clinical decision-making.
- MakaleAçık Erişim2025Advancing sea level anomaly modeling in the black sea with LSTM Auto-Encoders: A novel approach
Yavuzdoğan, A.; Tanir Kayıkçı, E.
Rising sea levels pose significant risks to coastal communities and ecosystems. Accurate modeling of sea level changes is crucial for effective environmental management and disaster mitigation. Machine learning methods are emerging as an important asset in improving sea level predictions and understanding the impacts of climate change. Especially, Long Short-Term Memory (LSTM) models have emerged as a powerful tool for sea level anomaly modeling, but there is an increasing need for more advanced models in this area. This study enhances existing methodologies by introducing a novel approach using an LSTM Auto-Encoder model, designed to compress input data into a lower-dimensional latent space before reconstructing it, thereby capturing complex temporal dependencies and anomalies effectively. We compared LSTM Auto-Encoder model performance with that of a Stacked LSTM network, which learns complex temporal patterns through multiple layers, and a traditional damped-persistence statistical model. Our results demonstrate that the LSTM Auto-Encoder model not only outperformed these models in predicting sea level anomalies across various lead times but also exhibited superior generalization capabilities across both satellite altimeter and in-situ data. These findings highlight the potential of the LSTM Auto-Encoder model as a powerful tool in coastal management and climate change studies, underscoring the critical role of advanced machine learning techniques in enhancing our predictive abilities and informing disaster preparedness strategies.
- MakaleAçık Erişim2025Lightweight Transformer Model for Agricultural Land Use and Land Cover Classification
Celik, Kemal
Observing agricultural land use via remote sensing images is essential for ensuring food security, estimating yields and planning efficient exports nonetheless precise classification continues to be difficult because of the varied and evolving characteristics of agricultural environments. This research aims to evaluate and optimize advanced deep learning architectures particularly Vision Transformer (ViT) models for agricultural land-use classification tasks. Specifically, we employed ViT-Base-16 and other lightweight models DeiT-Tiny and EfficientNet-B0 applying techniques such as model layer compression and advanced data augmentation CutMix and Cutout to achieve high accuracy while significantly reducing computational complexity. Evaluation was performed using three benchmark remote sensing datasets EuroSAT, NWPU-RESISC45 and SIRI-WHU which include diverse spatial resolutions and agricultural classes relevant for practical monitoring. Findings indicate that the optimized ViT algorithm is highly effective in recognizing global spatial connections, consistently achieving remarkable classification accuracy exceeding 99% on a newly assembled dataset containing around 200 samples of Google Earth imagery. Furthermore, for the first time in agricultural image classification compressing the ViT-Base model by pruning 50% of its layers significantly reduced complexity maintaining competitive accuracy 97.9% on SIRI-WHU. The resulting models are particularly suitable for deployment on devices with limited computational resources supporting real-world operational agricultural monitoring systems. This study emphasizes the revolutionary possibilities and practical use of optimized transformer-based models that offer scalable and efficient solutions specifically designed for precision agriculture applications.
- MakaleAçık Erişim2025A multi-stage fusion deep learning framework merging local patterns with attention-driven contextual dependencies for cancer detection
Catal Reis, Hatice; Turk, Veysel
Cancer is a severe threat to public health. Early diagnosis of disease is critical, but the lack of experts in this field, the personal assessment process, the clinical workload, and the high level of similarity in disease classes make it difficult. In recent years, deep learning-based artificial intelligence models have shown promise, with the potential to increase diagnosis speed and accuracy. These models attract attention with their automatic learning and adaptation capabilities. In this study, the deep learning-based PADBSRNet model and the PADBSRNet-Vision Transformer (ViT) hybrid method are proposed for the detection of brain tumors and skin and lung cancers. PADBSRNet is a comprehensive deep neural network architecture that integrates separable and traditional convolution layers, multiple attention mechanisms, bidirectional recurrent neural networks, and cross-connections/multi-stage feature fusion strategies. This architecture offers significant advantages in terms of effectively extracting local-global, contextual features and accurately modeling long-term dependencies in image classification tasks. The second proposed approach developed a hybrid method that combines the advantages of the PADBSRNet model and the ViT model. Experimental analysis on medical datasets such as the Figshare Brain Tumor Dataset, IQ-OTH/NCCD Dataset, and Skin Cancer: Malignant vs. Benign Dataset has evaluated the proposed models' performances. According to the experimental results, the PADBSRNet model has shown successful performance with 95.24 %, 99.55 %, and 88.61 % accuracy rates, respectively. The experimental findings show that the proposed deep learning model can effectively learn the complex relationships and hidden patterns of cancer disease, thus producing applicable and effective results in cancer diagnosis.
- MakaleAçık Erişim2025Feature Fusion Using Deep Learning Algorithms in Image Classification for Security Purposes by Random Weight Network
Kiran, Mustafa Servet; Seyfi, Gokhan; Yilmaz, Merve; Esme, Engin; Wang, Xizhao
Automated threat detection in X-ray security imagery is a critical yet challenging task, where conventional deep learning models often struggle with low accuracy and overfitting. This study addresses these limitations by introducing a novel framework based on feature fusion. The proposed method extracts features from multiple and diverse deep learning architectures and classifies them using a Random Weight Network (RWN), whose hyperparameters are optimized for maximum performance. The results show substantial improvements at each stage: while the best standalone deep learning model achieved a test accuracy of 83.55%, applying the RWN to a single feature set increased accuracy to 94.82%. Notably, the proposed feature fusion framework achieved a state-of-the-art test accuracy of 97.44%. These findings demonstrate that a modular approach combining multi-model feature fusion with an efficient classifier is a highly effective strategy for improving the accuracy and generalization capability of automated threat detection systems.
- MakaleAçık Erişim2025Enhancing mineral processing with deep learning: Automated quartz identification using thin section images
Külekçi, Gökhan; Hacıefendioğlu, Kemal; Başağa, Hasan Basri
The precise identification of quartz minerals is crucial in mineralogy and geology due to their widespread occurrence and industrial significance. Traditional methods of quartz identification in thin sections are labor-intensive and require significant expertise, often complicated by the coexistence of other minerals. This study presents a novel approach leveraging deep learning techniques combined with hyperspectral imaging to automate the identification process of quartz minerals. The utilizied four advanced deep learning models—PSPNet, U-Net, FPN, and LinkNet—has significant advancements in efficiency and accuracy. Among these models, PSPNet exhibited superior performance, achieving the highest intersection over union (IoU) scores and demonstrating exceptional reliability in segmenting quartz minerals, even in complex scenarios. The study involved a comprehensive dataset of 120 thin sections, encompassing 2470 hyperspectral images prepared from 20 rock samples. Expert-reviewed masks were used for model training, ensuring robust segmentation results. This automated approach not only expedites the recognition process but also enhances reliability, providing a valuable tool for geologists and advancing the field of mineralogical analysis.
- Yüksek Lisans TeziAçık Erişim2024Elastic stack ve makine öğrenmesi ile IoT güvenliği: Dos saldırılarına karşı akıllı savunma sistemi
Tuğrul Yağbasan
Teknolojinin gelişmesiyle IoT cihazların kullanımı artmaktadır. Bu cihazlar, iletişim kurma ve veri toplama gibi yeteneklerle endüstriyel, sağlık, ulaşım gibi birçok alanda kullanılmaktadır. Ancak, bu cihazların artan kullanımı, siber saldırıların da artmasına neden olmaktadır. Geçmişte Mirai botnet saldırıları gibi örnekler, IoT cihazlarının güvenliğinin ne kadar kritik olduğunu göstermiştir. Bu nedenle, çalışmada Elastic Stack, makine öğrenmesi ve derin öğrenme teknolojilerinin kullanımıyla IoT cihazlarının güvenliğini artırmaya yönelik akıllı savunma sistemi geliştirilmesi amaçlanmıştır. Çalışmada Bot-IoT ve Unsw-IoT-B datasetleri kullanılarak atakların belirlenmesi üzerine yoğunlaşılmıştır. Çalışmada makine öğrenmesi yöntemlerinden random forest, Decision Tree (Information Gain), Decision Tree (Gini Index), Naive Bayes teknikleri kullanılmıştır. Derin öğrenme yöntemlerinden Lstm, Rnn ve Svm kullanılarak datasetleri üzerinde atak ve atak tipleri tespiti gerçekleştirilmiştir. Tasarlanan modelde makine öğrenmesi yöntemleriyle %0.9998 lik başarım oranları yakalanmış olup derin öğrenme yöntemleriyle %0.98 lik doğruluk oranı elde edilmiştir. Bu modelle birlikte yapay zeka destekli saldırı tespit sistemlerinin IoT cihazlarının güvenliklerinde faydalı olacağını göstermektedir.
- BildiriAçık Erişim2024High-Accuracy Hand Segmentation for Palmprint Biometric Systems Improving Auto-Encoder Techniques
Yalcin, Kadir; Dogan, Ramazan Ozgur; Dogan, Hulya; Bingol, Ozkan
Biometric technologies have been widely adopted in various commercial products, ranging from security systems to personal devices, due to their exceptional reliability and user-friendliness. Among these, palmprint biometrics has gained increasing attention for its advantages, such as ease of use, a larger surface area, and the ability to capture more features compared to other biometric methods. A typical palmprint biometric system involves five key stages: (a) Image Acquisition, (b) Hand Segmentation, (c) Pattern Generation, (d) Feature Extraction, and (e) Verification/Identification. The overall success of the system depends on the effectiveness of each stage, with hand segmentation being crucial for identifying the most relevant features for accurate verification. This study aims to achieve high-accuracy biometric recognition by segmenting hands from images captured in uncontrolled environments with complex backgrounds. Accuracy (ACC), Intersection over Union (IoU), F1-Score, and Dice Coefficient are the four critical metrics used to assess the performance of segmentation techniques. The findings demonstrate that the U-Net + ViT and MA-Net + ResNet34 techniques outperform traditional auto-encoder methods in both quantitative performance metrics and visual analysis, leading to significantly improved hand segmentation for biometric systems.
- MakaleAçık Erişim2024Improving brain tumor classification with combined convolutional neural networks and transfer learning
İncir, Ramazan; Bozkurt, Ferhat
Brain tumors pose a serious threat, causing the deaths of thousands of people worldwide, and can lead to life-threatening consequences when not accurately diagnosed. The classification of brain tumors and the determination of correct treatment strategies are crucial, yet this process encounters various challenges. These tumors originate from different cell types and exhibit diversity in growth rates, histological features, and genetic structures. Some may show similarities at the microscopic level, complicating classification and making diagnosis and treatment challenging. The examination of brain MR images is a widely used method in diagnosing brain tumors. However, occasional misdiagnosis of tumors can lead to ineffective responses to treatments and reduced chances of survival for patients. Traditional machine learning classifiers require manually determined features, which is quite time-consuming. On the other hand, deep learning is highly effective in feature extraction and has recently been widely preferred in classification. In this context, the effectiveness of transfer learning architectures in brain tumor diagnosis was evaluated. Six different transfer learning architectures, including ResNet-50, MobileNet, VGG16, Inception-V3, DenseNet-121, and EfficientNetV2-M, were used in this study. A public MRI dataset was used for model validation and comparison with similar studies. To address the imbalance in the number of images among classes in the dataset, data augmentation techniques such as random rotation were applied during data preprocessing. Experiments revealed that the EfficientNetV2-M model outperformed other architectures with a 98.01% accuracy rate. Additionally, the study aimed to create a new model with more comprehensive feature extraction and generalization ability by combining the advantages of multiple models in different combinations. In this context, combinations of EfficientNetV2-M architecture with Inception-V3 and DenseNet-121 architectures were formed. Through these combinations, a concatenation-based EfficientNetV2-M + Inception-V3 model achieved a 98.41% accuracy value. It was observed that the proposed concatenation-based model outperformed advanced methods in improving medical imaging techniques and patient outcomes.
- MakaleAçık Erişim2024Advanced Tomato Disease Detection Using the Fusion of Multiple Deep-Learning and Meta-Learning Techniques
Catal Reis, Hatice
Automatic and early detection of diseases in crops is important for producers in terms of product yield. Methods are needed to diagnose the disease and select interventions for tomato leaf. This study aims to create a novel framework for automatically extracting tomato leaf disease. The recommended method automatically detects diseases in three classes: healthy, early, and late blight. This study developed a multi-deep and meta-learning fusion framework that automatically combines modern deep-learning algorithms and meta-classifiers to detect tomato leaf disease. The study includes CNN, transformer, and MLP-based models previously trained with the ImageNet dataset and voting classifier (VtC) methods, which are meta-classifiers. Hybrid approaches have been developed using the VtC method and deep learning algorithms such as CoAtNet, FasterNet, FastViT, WaveMLP, and DenseNet201. In addition to the proposed hybrid framework, different types of deep learning models based on CNN, transformer, MLP, transfer learning, and the dataset improved by the CLAHE technique and the original dataset were used to diagnose tomato leaf diseases. While CNN-based modern architectures include DenseNet201, MobileNet, LCNet, GhostNet, CoAtNet, ConvMixer, and FasterNet, transformer architectures include BeiT, FlexiViT, MaxViT, FastViT, and MLP-based architectures include MLP mixer and WaveMLP models. In the classification performed with original images, the proposed method produced incorrect results in only five images among 900 test data. In the classification made with CLAHE images, the proposed hybrid framework produced the most successful result in this study, producing incorrect results in only four images.
- MakaleYalnız künye2024Fast building detection using new feature sets derived from a very high-resolution image, digital elevation and surface model
Günen, Mehmet Akif
Detecting building rooftops with very high-resolution (VHR) images is an important issue in many fields, including disaster management, urban planning, and climate change research. Buildings with varying geometrical features are challenging to detect accurately from VHR image due to complicated image scenes containing spectrally similar objects, illumination, occlusions, viewing angles, and shadows. This study aims to detect building rooftops with high accuracy using a new framework that includes VHR image, visible band difference vegetation index, digital surface and elevation models, the terrain ruggedness and the topographic position index. Five distinct feature sets were generated in order of importance by exposing the ten related stacking features to a feature selection procedure using the maximum relevance minimum redundancy method. Then, Auto-Encoder, k-NN, decision tree, RUSBoost, and random forest machine learning algorithms were utilized for binary classification. Random forest yielded the highest accuracy (97.2% F-score, 98.72% accuracy) when all features (F10) were used, while decision tree was the least successful (59.16% F-score, 83.56% accuracy) for RGB feature set (FRGB). It was revealed that classification of F10 with random forest increased F-score by about 23% compared to classification with FRGB. Additionally, McNemar’s tests showed no statistically significant difference between random forest vs k-NN and decision tree vs RUSBoost.
