Serkan Ozbay

Accurate and efficient detection of defects during the manufacturing process of photovoltaic (PV) panels is critical for ensuring product quality and operational reliability. However, developing lightweight deep learning models capable of handling diverse defect types remains a significant challenge. To address this issue, this study introduces a comprehensive 10-class electroluminescence (EL) dataset to the literature, which includes the Striation Rings class, a defect type rarely represented in publicly available datasets. Crucially, to overcome class imbalances and significantly boost class-specific detection performance, a stable diffusion-based data augmentation strategy is implemented to generate highly realistic synthetic defect samples for underrepresented classes. Furthermore, this study proposes a lightweight and enhanced You Only Look Once (YOLO)11-based architecture specifically designed for superior defect detection. The proposed model integrates a spatial gated C3 transformer module and a fusion module into the base YOLO11s architecture to enhance feature extraction and better capture critical spatial information. Additionally, Wise-IoU is used as the loss function to optimize bounding box regression. In order to assess its effectiveness, the proposed model is evaluated against a comprehensive suite of state-of-the-art object detection frameworks, encompassing successive YOLO generations, Faster R-CNN, and RF-DETR. Experimental evaluations reveal that the proposed model surpasses all competing methods by a substantial margin. The model attains an mAP@0.5 score of 0.9876, yielding an 8.81% improvement over the baseline YOLO11s architecture. In addition, extensive cross-dataset experiments conducted on the PVEL-AD benchmark further validate the model’s generalization capability. The results consistently demonstrate superior defect detection performance, highlighting the effectiveness, robustness, and reliability of the proposed framework for real-world photovoltaic defect inspection applications.

The use of textile-based sensors has increased significantly in recent years, and wearable textile-based sensors have become superior interfaces for bio-signal sensing. Surface electromyography (sEMG), a widely used method for recording the electrical activity of muscles, has been used with textile-based electrodes. However, sEMG signals can be corrupted by various noises, interference and artifacts. In this study, an effective noise reduction algorithm has been developed on sEMG signals obtained from textile-based electrodes. The algorithm introduces a novel hybrid scheme which includes wavelet thresholding to wavelet neural network. The performance of the proposed algorithm is analyzed with real data and compared with existing algorithms. It is shown that the proposed algorithm achieves remarkable performance in noise reduction and the results obtained with the proposed algorithm are superior to the results obtained with the state-of-the-art existing algorithms. The optimum performance results of the proposed algorithm are observed as a mean square error (MSE) value of 0.0014, a root means square error (RMSE) of 0.0374, and a correlation coefficient (r) value of 0.97 with sEMG signals obtained by textile-based electrodes. It is seen that the proposed algorithm significantly improves the quality of sEMG signals for wearable applications thus it is believed to be a potential candidate on the signal acquisition of wearable technologies.

This study investigates the impact of response time on ability estimation within an Item Response Theory (IRT) framework, introducing time-sensitive formulations to enhance student assessment accuracy. Seven models were evaluated, including standard 1PL-IRT and six response-time-adjusted variants: TP-IRT, STP-IRT, TWD-IRT, NRT-IRT, DTA-IRT, and ART-IRT. Three optimization techniques—Maximum Likelihood Estimation (MLE), full parameter optimization, and K-fold Cross-Validation (CV)—were employed to assess model performance. Empirical validation was conducted using data from 150 students solving 30 mathematics items on the “TestYourself” platform, integrating response accuracy and timing metrics. Student abilities (θ), item difficulties (b), and time–effect parameters (λ) were estimated using the L-BFGS-B algorithm to ensure numerical stability. The results indicate that subtractive models, particularly DTA-IRT, achieved the lowest AIC/BIC values, highest AUC, and improved parameter stability, confirming their effectiveness in penalizing excessive response times without disproportionately affecting moderate-speed students. In contrast, multiplicative models (TWD-IRT, ART-IRT) exhibited higher variability, weaker generalizability, and increased instability, raising concerns about their applicability in adaptive testing. K-fold CV further validated the robustness of subtractive models, emphasizing their suitability for real-world assessments. These findings highlight the importance of incorporating response time as an additive factor to improve ability estimation while maintaining fairness and interpretability. Future research should explore multidimensional IRT extensions, behavioral response–time analysis, and adaptive testing environments that dynamically adjust item difficulty based on response behavior.

Trafik işaretleri, karayolunda seyahat eden sürücülere yol kısıtlamaları açısından uyarıda bulunmak amacıyla karayollarına yerleştirilmektedir. Bu işaretlerin doğru bir şekilde algılanması ve trafik işaretinin gerektirdiği kısıtlamaya uyulması, sürüş güvenliği açısından önemlidir. Son yıllarda, derin öğrenme algoritmalarının nesne sınıflandırılmasında ve tespitinde başarılı olduğu birçok çalışma ile gösterilmiştir. Bu çalışmada, derin öğrenme tabanlı “You Only Look Once” (YOLO) algoritmaları trafik işaretleri tespiti açısından karşılaştırılmıştır. İlk olarak 877 görüntüden oluşan dört sınıflı trafik işaretleri veri seti elde edilmiştir. Daha sonra YOLOv5, YOLOv8 ve YOLOv9 algoritmaları trafik işareti tanıması açısından incelenmiştir. Deneysel çalışmalarda, tespit algoritmalarının performanslarını değerlendirmek amacıyla duyarlılık, kesinlik, f1 skor ve mAP performans değerlendirme kriterleri açısından incelenmiştir. Elde edilen deneysel sonuçlara göre YOLOv9’un Duyarlılık metriği %90.8, mAP@0.5 metriği %93.1 ve mAP@0.5:0.95 metriği %77.7 olarak hesaplanmıştır. Bu sonuçlar YOLOv9 algoritmasının en iyi trafik işareti tespit algoritması olduğunu doğrulamaktadır.

Backpropagation is one of the most widely used algorithms for training feedforward deep neural networks. The algorithm requires a differentiable activation function and it performs computations of the gradient proceeding backwards through the feedforward deep neural network from the last layer through to the first layer. In order to calculate the gradient at a specific layer, the gradients of all layers are combined via the chain rule of calculus. One of the biggest disadvantages of the backpropagation is that it requires a large amount of training time. To overcome this issue, this paper proposes a modified backpropagation algorithm with multiplicative calculus. Multiplicative calculus provides an alternative to the classical calculus and it defines new kinds of derivative and integral forms in multiplicative form rather than addition and subtraction forms. The performance analyzes are discussed in various case studies and the results are given comparatively with classical backpropagation algorithm. It is found that the proposed modified backpropagation algorithm converges in less time to the solution and thus provides fast training in the given case studies. It is also shown that the proposed algorithm avoids the local minima problem.

Visual object trackers based on deep neural networks have attained state-of-the-art performance in recent years. Despite the outstanding accuracy gained by deep layers, however, they also demand high computational cost and energy consumption in order to operate in real-time, making them inadequate for edge and latency-sensitive applications. In this paper, we propose an edge computing-friendly Siamese-based visual object tracker. This work concentrates on increasing the tracking speed by reducing computations through integration of side exit branches into the network, as well as skipping the multi-scale search for some frames. By employing exit branches, the tracker is capable of obtaining the result of easy samples from early layers once the criteria are satisfied. The network is trained offline to optimize a joint function that is composed of the weighted loss functions of all exit branches. During inference, the score map is derived from the network and determines the new object location, whereas multi-scale testing can identify scale updates which is only applied under specific conditions. Our proposed tracker deploys an adaptive scale search over two scales that runs at 247.5 FPS on GPU and 37.1 FPS on CPU providing a 2.5x faster rate of processing speed compared to SiamFC, with an acceptable amount of accuracy loss, especially when compared to the significant speed gain and gains in computational efficiency.

Due to the rapid increase for novel antenna designs to satisfy the modern wireless technologies, different techniques are proposed to improve their performance. For this purpose, this paper presents a novel technique to design a miniaturized microstrip antenna structure based on metamaterial (MTM) transmission line defected patch. The proposed antenna is introduced as part of wearable systems with minimum back radiation toward the human body. Therefore, the antenna is constructed of a closed loop linked to an interdigital capacitor to magnify the electric field fringing at the patch core. The proposed antenna when printed on a FR-4 substrate at 0.6GHz in the first mode shows S11 impedance matching less than -10dB. The proposed antenna performance is evaluated numerically based on CST MWS in terms of radiation pattern. A neural network is invoked to optimize the proposed antenna performance at the required frequency band. Finally, the proposed antenna performance is validated using another software package based on HFSS simulator to provide good agreement with percentage of error less than 4.4%.

İnsan ve bilgisayar etkileşimi konusu, son zamanlarda giderek daha popüler hale gelen konulardan biridir. Bilgisayar ile insan yüz ifadesi ve duygu analizi ilginç olduğu kadar karmaşık sorunlardan biridir. Bu çalışmada insan görüntüleri üzerinden duygu analizi yapılmıştır. Çalışmada mutlu, kızgın, üzgün, şaşırmış ve nötr olmak üzere 5 farklı duygu durumu analiz edilmiştir. Önerilen algoritma temelde 3 adımdan oluşmaktadır. Birincisi, SVM modeli için gerekli görüntülerin ön işlemesidir. İkincisi, duygu analizi yapabilen SVM modelinin oluşturulmasıdır. Son adım,yüz ifadelerin ilgili duygu sınıfa yönlendirilmesidir. Bu çalışmada, JAFFE veri seti ve Google'da bulunan birçok görsel kullanılmıştır. Çalışma için belirlenen 5 farklı duygu için başarı oranları % 80 ile % 100 arasında değişmektedir.