Muhammet Fatih Aslan

Automated fault detection in wind turbine blades is critical for ensuring the reliability and operational efficiency of wind energy systems. This study presents a systematic comparative analysis of five state-of-the-art deep learning architectures for binary fault classification (healthy versus faulty) on the CAI-SWTB dataset, comprising 6,000 RGB images of small wind turbine blades. The evaluated architectures span three distinct design paradigms: a classical Convolutional Neural Network (CNN) (ResNet-50), a modern CNN (EfficientNetV2-S), and three Vision Transformer (ViT)-based models (ViT-B/16, Data-efficient Image Transformer (DeiT)-S/16, and Swin-Tiny). All models were trained using a two-stage transfer learning protocol with ImageNet-pretrained weights, employing the AdamW optimizer and a cosine annealing learning rate schedule. EfficientNetV2-S achieved the highest classification performance with 99.75% accuracy, followed by ResNet-50 at 99.42%. Among the transformer-based models, Swin-Tiny outperformed both ViT-B/16 (65.33%) and DeiT-S/16 (82.75%), achieving 88.25% accuracy. Grad-CAM analysis confirmed that the best-performing models correctly localize structural defect regions in blade images, supporting their interpretability and suitability for real-world inspection applications.

Recent advances in medical image segmentation have introduced a wide spectrum of deep learning paradigms, including 2D/3D convolutional neural networks (CNNs), transformer-based architectures, vision-language models (VLMs), prompt-driven foundation models such as Segment Anything Model (SAM), and diffusion-based approaches. Although these methods have demonstrated remarkable performance across MRI, CT, PET, ultrasound, and endoscopic imaging, the rapid proliferation of architectures has created methodological uncertainty regarding optimal model selection under varying clinical and data constraints. Existing surveys primarily focus on architectural categorization, yet provide limited guidance for decision-oriented model selection. This study presents a comprehensive and decision-guided survey that systematically analyzes segmentation paradigms across imaging modalities, task types, dataset characteristics, and evaluation protocols. Beyond taxonomy, we propose a practical model selection framework that links clinical scenarios, such as small lesion detection, multi-organ 3D segmentation, limited-data regimes, and domain shift, to appropriate segmentation strategies. Furthermore, robustness, generalization, annotation variability, and benchmarking reproducibility are critically examined. By integrating architectural taxonomy, cross-modal comparative analysis, and a structured decision framework, this work provides a clinically oriented roadmap for selecting segmentation methods and highlights future research directions toward reliable and reproducible medical AI systems.

Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. However, their application in safety-critical domains such as elevator emergency response requires domain-specific knowledge and reliable performance, particularly when deployed as offline edge AI systems with constrained computational resources. This study presents fine-tuning of the Gemma-2-9B-Instruct model using Quantized Low-Rank Adaptation (QLoRA) for Turkish elevator emergency scenarios, targeting deployment on embedded touchscreen control panels inside elevator cabins. Unlike cloud-based systems leveraging internet connectivity for retrieval, our edge deployment operates entirely offline, making fine-tuning essential for encoding domain knowledge directly into model parameters. We developed a specialized Turkish dataset containing 1,155 question-answer pairs covering diverse emergency situations. Our evaluation demonstrates significant improvements: ROUGE-1 scores increased from 0.259 to 0.317 (22.49%), BLEU improved by 218.11%, and hallucination rates reduced from 54% to 22% while achieving 49% faster inference. The resulting 4.8GB quantized model runs entirely on embedded hardware without network dependencies, providing immediate, reliable emergency guidance. These results validate parameter-efficient fine-tuning for safety-critical edge AI applications.

Vision Transformers (ViTs) have achieved strong performance in computer vision but suffer from limited inductive bias, high data requirements, and reduced sensitivity to high-frequency visual details. To address these limitations, Frequency-Domain ViTs (FD-ViTs) incorporate spectral representations—such as Fourier, wavelet, and discrete cosine transforms—into the Transformer pipeline to improve feature expressiveness and robustness. This survey provides a systematic review of FD-ViT architectures and introduces a unified taxonomy based on spectral transformation type, integration level, and computational characteristics. We summarize empirical findings across image classification, image restoration, and domain-specific applications, including medical imaging and remote sensing, highlighting consistent performance patterns and task-dependent trade-offs. Our analysis shows that frequency-domain integration yields modest, context-dependent gains in large-scale classification, while offering more consistent advantages in frequency-sensitive tasks such as image restoration and noise-robust visual analysis. We further discuss key open challenges, including spectral aliasing, phase information loss, evaluation inconsistency, and deployment efficiency, and outline emerging directions toward dynamic spectral operators, multimodal integration, and hardware-aware designs. To the best of our knowledge, this work constitutes the first systematic survey that consolidates the growing body of research on FD-ViT, providing a structured conceptual and methodological reference for future studies on spectral representations in Transformer-based visual learning.

In this review, the details and developments of few‐shot learning (FSL) techniques in different remote sensing (RS) studies including change monitoring, disaster management, urban monitoring, and agriculture are discussed in detail. Furthermore, a categorization is made by dividing FSL methods into three categories (metric‐based, optimization‐based, and transfer learning approaches) and considering hybrid approaches. Special attention is given to episodic training and meta‐learning approaches that provide rapid adaptation to new classes with minimal examples. Furthermore, the integration of explainable artificial intelligence (XAI) and its real‐time application capabilities are discussed. Important issues such as domain shift, class imbalance, and high dimensionality are discussed. Recent refinements such as task‐level learning, data augmentation, and multimodal integration are examined. Finally, a coherent framework is suggested for further studies and practical FSL applications in the context of RS. As a result, it provides a more comprehensive perspective than previous reviews. This review aimed to guide future research in the integration of FSL with RS applications by analyzing the existing literature and pointing out important research gaps.

Since each of the elevator motherboards produced in Turkey today uses its own special communication protocol, the panels must be adapted separately to each system, and this leads to serious time and cost loss in the production process. For this purpose, this study aims to ensure that elevator control cards developed by different manufacturers work in harmony with common call and display panels. In this context, a call and display panel compatible with the CANopen Lift (CIA-417) profile, one of the international standards, has been designed. The embedded software infrastructure was created with the CANopenNode open-source library and the system's error analysis was performed with the CAN Analyzer. In addition, the sustainability and usability of the system was increased with the architecture that allows software updates and parameter changes to be made remotely. The prototype developed at the end of the project was tested in pilot elevators and successful results were obtained. With this study, a domestic solution that will reduce Turkey's dependence on imports has been presented. In addition, the way will be paved for CIA-417 compatible panel production.

In this study, a dual frequency supported Radio Frequency Identification (RFID) card reader device that can operate at both 125 kHz and 13.56 MHz is developed for use in access control systems. Most of the existing systems operate at a single frequency band and therefore offer limited card compatibility. But, the proposed design supports multiple RFID standards and offers a more flexible and user-friendly solution. The reader functioning at 125 kHz can successfully detect cards based on the EM4100 protocol, whereas at 13.56 MHz, it effectively communicates with MIFARE Classic, MIFARE Ultralight, and Near Field Communication (NFC)-compatible cards. In the hardware design, two RFID modules have been optimized and integrated under the control of a single microcontroller. The system is capable of automatically detecting the card's frequency, switching to the appropriate protocol, and performing secure identity verification processes. Additionally, the device has been designed to be resistant to electromagnetic interference and capable of operating inside a metal enclosure, taking into account harsh environmental conditions. Thus, it can also exhibit reliable performance in areas such as industrial facilities, parking systems and dense metal environments. Thanks to this innovative approach, not only has compatibility with existing card infrastructures been achieved, but also an important step has been taken for long-lasting and secure access control systems.

Background: Gastrointestinal (GI) disorders present significant healthcare challenges, requiring rapid, accurate, and effective diagnostic methods to improve treatment outcomes and prevent complications. Wireless capsule endoscopy (WCE) is an effective tool for diagnosing GI abnormalities; however, precisely identifying diverse lesions with similar visual patterns remains difficult. Methods: Many existing computer-aided diagnostic (CAD) systems rely on manually crafted features or single deep learning (DL) models, which often fail to capture the complex and varied characteristics of GI diseases. In this study, we proposed “EndoNet,” a multi-stage hybrid DL framework for eight-class GI disease classification using WCE images. Features were extracted from two different layers of three pre-trained convolutional neural networks (CNNs) (Inception, Xception, ResNet101), with both inter-layer and inter-model feature fusion performed. Dimensionality reduction was achieved using Non-Negative Matrix Factorization (NNMF), followed by selection of the most informative features via the Minimum Redundancy Maximum Relevance (mRMR) method. Results: Two datasets were used to evaluate the performance of EndoNer, including Kvasir v2 and HyperKvasir. Classification using seven different Machine Learning algorithms achieved a maximum accuracy of 97.8% and 98.4% for Kvasir v2 and HyperKvasir datasets, respectively. Conclusions: By integrating transfer learning with feature engineering, dimensionality reduction, and feature selection, EndoNet provides high accuracy, flexibility, and interpretability. This framework offers a powerful and generalizable artificial intelligence solution suitable for clinical decision support systems.

Derinin insan hayatında ilk çağlardan beri var olduğu tahmin edilmektedir. Nitekim ilk başlarda sadece soğuktan ve rüzgârdan korunmak için kullanıldığı düşünülürken günümüzde deri, dayanıklılığı ve estetikliği sebebiyle moda, mobilya ve otomotiv gibi devasa sektörlerde çok geniş bir kullanım alanına sahiptir. Küresel deri pazarı her geçen yıl daha da büyümektedir. Bu durum deri üretiminde akıllı yaklaşımların önemini her geçen gün artırmaktadır. Tabakhanelerde deri ürünlerindeki yüzey kusurlarını tespit edebilmek için tecrübeli insan denetçilerin kullanılması uzun zamandır süregelen bir uygulamadır. Ancak bu yöntem, yorucu, zaman alıcı, hatalara açık ve kişinin tecrübesine göre değişkendir. Kaliteli deri ürünlerinde hata payının düşük olması ekonomik olarak ciddi öneme sahiptir. Sektördeki insan hatalarından kurtulmak ve verilen kararları nesnelleştirebilmek adına deri yüzeyindeki kusurların otomatik tespit edilebilmesi ihtiyacı ortaya çıkmıştır. Bu çalışmada hayvan derisinin kusurlarını tespit edebilmek amacı ile makine öğrenmesi tekniklerini kullanan çalışmalar hakkında literatür taraması yapılmıştır. Görüntü analizi ve makine öğrenmesi yöntemleri ile deri yüzeylerindeki kusurları tespit etme yöntemlerinin detaylı bir incelemesi yapılmıştır. Bu alanda gelecekte yapılacak çalışmaları teşvik etmek amacı ile zorluklar ve gelişim evreleri sunulmuştur.