Selda Güney

Monitoring of Methyl Ethyl Ketone (MEK) is essential for industrial safety and biomedical diagnostics, serving as a reliable biomarker for metabolic disorders. Conventional detection techniques require bulky, expensive instrumentation, limiting portable use. In this study, a low-cost and disposable electrochemical sensor was fabricated using a Pencil Graphite Electrode (PGE) modified with a porous Hydroxyapatite-like (HAp-like) interlayer and Copper Oxide (CuO). The HAp-like scaffold minimized the copper agglomeration and increased the electroactive surface area, with SEM and EDX confirming homogeneous CuO distribution. Toward MEK, the sensor exhibited a “signal-off” response, systematically suppressing the anodic peak current. This is consistent with a steric-hindrance mechanism where analyte adsorption blocks interfacial electron transfer. Evaluated via square-wave voltammetry (SWV) in phosphate buffer (PBS, pH 10.0), the sensor demonstrated dual linear ranges (0.1-5.0 mM and 5.0-50 mM) with a 1.04 mM limit of detection. The sensor demonstrated consistent reproducibility with an overall relative standard deviation (RSD) of 2.83% (n=15) and an inter-day RSD of 1.94%, while retaining 107.2% of its initial response after a 14-day storage period. Validated in PBS pH 10.0, this CuO/HAp-like/PGE platform is presented as a proof-of-concept for liquid-phase MEK detection, establishing a foundation for future monitoring under simulated physiological conditions.

With the rise of Industry 4.0, the growing automation of production processes has necessitated the development of more intelligent and autonomous quality control systems. In this context, image processing-based decision support systems (DSS) provide a highly accurate, fast, and consistent approach to performing quality control on production lines. These systems use high-resolution cameras to capture visual data from products. This data is then preprocessed through steps such as noise reduction, contrast enhancement, and thresholding to prepare it for analysis. Surface defects, color deviations, dimensional anomalies, and deformations are detected using techniques such as mathematical morphology, edge detection, and segmentation. The extracted features are then analyzed using either traditional classification algorithms or deep learning models such as Convolutional Neural Network (CNN), U-Net, or ResNet, enabling the classification of products as “defective” or “non-defective”. For this study, CNNs, a widely used discrimination-based method in image processing, were selected. CNN algorithms take an input image and learn to identify and distinguish various features within it. Based on this, a model was developed that uses the best-performing architecture. An analysis of a dataset containing over 30 types of defects, including beading and tearing, showed that the model achieved an accuracy rate of 82.8% in detecting weaving defects. Accordingly, a quality control system based on the proposed image processing model has been developed. Integrating this system into the company's production environment and evaluating its real-time performance will be the focus of future work. This integration is expected to support the development of Industry 4.0 in Türkiye 's home textile sector and strengthen international competitiveness by offering a model image processing-based quality control system that can serve as a reference for other manufacturers in the country.

Modern military intelligence systems are advancing rapidly, particularly in the classification of combat vehicles in battlefield environments. Accurately identifying the capabilities and vulnerabilities of these vehicles is crucial for developing effective tactics and operational strategies. The integration of artificial intelligence, specifically through image processing, pattern recognition, and deep learning techniques, has greatly enhanced the feasibility of military vehicle classification. This study employs the You Only Look Once (YOLO) algorithm, particularly the YOLOv8 model, which is optimized for mobile applications and has shown high performance in evaluations. Enhancements include the incorporation of a squeeze and excitation (SE) block, improving detection accuracy. The research emphasizes the importance of detecting detailed target characteristics using images collected from electro-optical systems and aims to bolster decision support processes in target management systems. The development of a deep learning-based Target Detection and Identification (TDI) system is outlined, encompassing specialized targeted data acquisition, pre-processing, image partitioning, feature extraction, classification, and recommendation generation through decision support matrices. This model significantly improves image processing and object detection capabilities. Notably, it facilitates passive identification of combat vehicles using existing electro-optical devices without the need for additional hardware, while estimating distances discreetly. The system delivers real-time predictions, achieving a mean Average Precision (mAP) of 88.38%, thereby facilitating informed decision-making and risk reduction in military operations.

Object classification in radar signal processing is one of the most challenging problems for similar featured targets. This challenge arises due to the limited accuracy of traditional classification methods, such as those based on radar cross section (RCS), speed, and phase differences. In the context of drone-bird classification, the sizes of drones and birds result in similar RCS signatures. Consequently, traditional classification algorithms are not effective in this context. This study addresses this issue by analyzing the micro-Doppler signatures of In-phase Quadrature (IQ) data obtained from Frequency Modulated Continuous Wave radar. The IQ data are then fed into pre-trained neural networks, classical machine learning algorithms, and convolutional neural networks (CNNs) with varying architectures to classify drones and birds separately. As a result of this study, high success rates were obtained with CNNs, Pre-Trained Networks (VGG16, VGG19, Alexnet, Googlenet, and Resnet50), and a combination of CNN algorithms with traditional machine learning algorithms like Support Vector Machine, Tree Methods, etc. Comparison results show that the combination of CNNs with traditional machine learning algorithms gives promising results.

In this study, a hybrid sensor-based electronic nose circuit was developed using eight metal-oxide semiconductors and 14 quartz crystal microbalance gas sensors. This study included 100 participants: 60 individuals diagnosed with lung cancer, 20 healthy nonsmokers, and 20 healthy smokers. A total of 338 experiments were performed using breath samples throughout this study. In the classification phase of the obtained data, in addition to traditional classification algorithms, such as decision trees, support vector machines, k-nearest neighbors, and random forests, the fuzzy logic method supported by the optimization algorithm was also used. While the data were classified using the fuzzy logic method, the parameters of the membership functions were optimized using a nature-inspired optimization algorithm. In addition, principal component analysis and linear discriminant analysis were used to determine the effects of dimension-reduction algorithms. As a result of all the operations performed, the highest classification accuracy of 94.58% was achieved using traditional classification algorithms, whereas the data were classified with 97.93% accuracy using the fuzzy logic method optimized with optimization algorithms inspired by nature.

In this study, a hybrid sensor-based electronic nose circuit was developed using eight metal-oxide semiconductors and 14 quartz crystal microbalance gas sensors. The study included 100 participants: 60 individuals diagnosed with lung cancer, 20 healthy nonsmokers, and 20 healthy smokers. A total of 338 experiments were performed using the breath samples throughout the study. In the classification phase of the obtained data, in addition to traditional classification algorithms, such as decision trees, support vector machines, k-nearest neighbors, and random forests, the fuzzy logic method supported by the optimization algorithm was also used. While the data were classified using the fuzzy logic method, the parameters of the membership functions were optimized using a nature-inspired optimization algorithm. In addition, principal component analysis and linear discriminant analysis were used to determine the effects of dimension-reduction algorithms. As a result of all the operations performed, the highest classification accuracy of 94.58% was achieved using traditional classification algorithms, whereas the data were classified with 97.93% accuracy using the fuzzy logic method optimized with optimization algorithms inspired by nature.

Swift process in technology and widespread availability of low-cost internet have led to a substantial rise in data volume in remote sensing, especially for high-resolution and very-high resolution images. Still, these images contain more complex information, and it is not appropriate to analyze the images using a solitary scene-level label while ignoring the distinct features provided by other labels in the images. In multi-label image classification applications, multiple labels are assigned to an image, reflecting various objects or features present in the scene. The classification of these images is critically important for monitoring environmental changes over large geographical areas, disaster management, urban planning, agriculture and forestry management, natural resource conservation, and military intelligence. Nowadays, many methods are used in such image classification problems, primarily deep learning algorithms. However, current deep learning approaches for multi-label remote sensing images often struggle to capture both local fine-grained details and global contextual relationships simultaneously, leaving a gap for models that can efficiently integrate these complementary representations. In this study, advanced neural networks are explored and evaluated for Multi-label AID dataset which contains 3000 images and 17 different labels; AlexNet, VGG16, DenseNet-201, Inception-v3 and ConvNeXt as the CNN models, ViT, SwinT as transformer models and MaxViT as the hybrid model that initially contains both CNN and transformer network. OneCycleLR as scheduler and AsymmetricLoss (ASL) as loss function are employed for each model to systematically evaluate their impact on model performance. MaxViT was chosen because its multi-scale window-based attention can jointly model local and global dependencies, making it particularly suitable for the complex spatial patterns in remote-sensing imagery compared with other hybrid architectures. The window-based MaxViT algorithm, which has not been previously applied to the Multi-label AID dataset in the current literature, has been evaluated. This constitutes the first application of MaxViT to this dataset and provides a novel benchmark for multi-label remote-sensing classification. This algorithm has demonstrated superior performance on this dataset, significantly outperforming existing models and setting a new benchmark with an mAP of 84.98%.

Breast cancer has a high incidence and mortality rate in the female population. Mammography is the most reliable method for early and accurate diagnosis of breast cancer. Automated detection and classification of breast masses on mammograms is a challenging task and is essential to assist radiologists in accurately diagnosing breast masses. The aim of this study is to develop a Computer-Aided Diagnosis (CAD) system based on You Look Only Once (YOLO) for identifying breast masses and classifying them as benign or malignant. We propose a YOLOv5-CAD framework that uses a transfer learning approach. Two datasets, CBIS-DDSM and VinDr-Mammo, are utilized for training from scratch. The model weights and parameters are subsequently transferred and fine-tuned onto the smaller INBreast dataset. Furthermore, an analysis is conducted to assess the impact of various data augmentation techniques during the training phase on enhancing model performance. The proposed framework demonstrates encouraging fivefold cross-validation evaluation results. To conclude, transfer learning from CBIS-DDSM achieves 0.843 mAP, precision of 0.855, recall of 0.774, while transfer learning from VinDr- Mammo reaches 0.84 mAP, precision of 0.829, recall of 0.787. Furthermore, the performance of the two fine-tuned models was tested on both the MIAS dataset and the private dataset from Başkent University Ankara Hospital. Such promising performance could be useful for the CAD frameworks being developed to support radiologists as a second opinion reader for the detection and classification of breast masses.

Günümüzde trafik güvenliği, gelişmiş sürücü destek sistemlerini (ADAS) önemli ölçüde etkilenmektedir. Sürücüler için önemli bilgiler taşıyan trafik işaretleri çoğu zaman sürücüler tarafından fark edilemez ve bunun sonucunda da çeşitli kazalar meydana gelir. Bu sebeplerle ADAS’ın içinde yer alan trafik işaretlerini algılayan ve sınıflandıran sistemler mevcuttur. Derin öğrenme teknikleri, özellikle de konvolüsyonel sinir ağları üzerinde yapılan çalışmalar, bu alanda önemli ilerlemelere yol açmıştır. Son yıllarda ResNet50, GoogleNet, SqueezNet ve AlexNet gibi birçok derin öğrenme algoritmaları, trafik işaretlerini daha doğru bir şekilde sınıflandırmaya yardımcı olmak için kullanılmıştır; ancak bu algoritmaların gerçek zamanlı bir sistemde iyi bir performans gösterebileceği konusu tartışmalıdır. Bu çalışmada derin öğrenme algoritmaları transfer öğrenim metodu ile kullanılarak sınıflandırma işlemleri gerçekleştirilmiş ve bu algoritmaların gerçek zamanlı bir sistemde performansları test edilmiştir. Bu çalışmada veri seti olarak, 43 sınıftan oluşan ve 39209 trafik işareti görüntüsü içeren German Traffic Sign Recognition Benchmark (GTSRB) kullanılmıştır. Öncelikle bu algoritmaların hiper parametreleri sistematik bir yöntemle belirlenmiştir. Daha sonra tasarlanan bir kullanıcı arayüzüyle gerçek zamanlı deneyler yapılmıştır. Çalışmada kullanılan mimarilerden ResNet50, %93,49 doğruluk oranı ile gerçek zamanlı bir sistemde en iyi performansı verdiği gözlemlenmiştir. Sonuçta, ResNet50 derin öğrenme algoritması, trafik işaretinin gerçek zamanlı bir sistemde algılanması alanındaki ileri çalışmalar için değerlendirilebilir.

Geçmişten günümüze hava tahmini insanlık için önem arz etmektedir. Hava tahmininin hassas gerçekleştirilebilmesi sel, tsunami vb. doğal afetlere karşı önlemler alınarak oluşacak olumsuz etkileri en düşük seviyeye indirmeyi sağlayabilmektedir. Bu çalışma kapsamında radyosonde verilerini kullanarak hava durumu kestirimi yapılmaktadır. Bu kestirimde en yüksek ve en düşük sıcaklık tahmini yapılmaktadır. Makine Öğrenmesi Algoritmaları kullanarak kestirim gerçekleştirilmiştir. Daha önce literatürde bulunan sıcaklık tahmini çalışmalardan farklı olarak 3 yıllık Radyosonde rasat verileri kullanılmıştır. Bu sayede yerden 40 km yüksekliğe kadar 1mbar aralıklarla ölçülmüş veriler ile atmosfer, literatürdeki diğer çalışmalara göre çok daha hassas olarak modellenmiştir. Bu modelde ertesi güne ait en yüksek ve en düşük sıcaklık değerleri kestirilmiştir. Bu aşamada normalizasyon ve öznitelik çıkarma veya seçmenin sonuçlara etkileri analiz edilerek tahmin için en uygun model belirlenmiştir. MATLAB ortamında gerçekleştirilen yazılım ile faklı regresyon yöntemleri karşılaştırılmıştır. Bu analizler sonucunda Gauss Süreci Regresyonu yöntemini kullanarak 1,2 Ortalama Karekök Sapması ile ertesi güne ait en yüksek sıcaklık tahmini en yüksek doğrulukla elde edilmiştir. Aynı yöntem kullanarak 2,4 Ortalama Karekök Sapması oranı ile en düşük sıcaklık tahmini yapılmıştır. Sonuçlar literatürdeki çalışmalardan daha başarılı sıcaklı tahmini yapıldığını göstermektedir.