HCHüseyin Canbolat

Hüseyin Canbolat

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Hüseyin Canbolat has a total of 16 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A multi-stage convolutional and self-attention architecture for high-precision sign language gesture recognition

    by Ahmed Kasapbaşı, Hüseyin Canbolat

    This paper presents a novel composite architecture for high-precision sign language gesture recognition referred to as ViTSLA. The model integrates convolutional inductive bias with transformer-based global attention to improve predictive accuracy and computational efficiency. Extensive experiments are conducted on three datasets using five-fold cross-validation, ablation analysis, and comparative evaluation against standard vision transformer (ViT) variants. Experimental results demonstrate consistently high classification accuracy exceeding 99.5% with minimal variance across folds, confirming robustness and balanced class-wise performance. Beyond predictive performance, this work addresses the computational demands of training and deploying hybrid deep architectures. Transformer-based attention mechanisms and high-resolution tokenization increase memory consumption and computational complexity. Therefore, efficient training requires parallel processing, GPU acceleration, and high-performance computing (HPC) environments to handle large-scale tensor operations and multi-fold cross-validation experiments. Furthermore, real-time inference capability is maintained at over 300 FPS, which is essential for deployment in latency-sensitive applications such as interactive systems and intelligent human–computer interfaces. Distributed computation further enables scalable training and stable convergence across datasets. Moreover, the low mean latency, stable P95 behavior, and high throughput confirm real-time deployment capability. These results demonstrate that the proposed architecture achieves high predictive performance while maintaining computational characteristics aligned with supercomputing-driven intelligent vision systems. The developed model achieves a favorable balance between accuracy, parameter efficiency, and computational cost. The results demonstrate that hybrid convolutional-attention architectures benefit from HPC-enabled parallelism, leading to scalable optimization and deployment-ready performance. These findings establish the relevance of the proposed approach within the domain of supercomputing-driven intelligent vision systems.

  • POST-QUANTUM CRYPTOGRAPHY SABER IN A HYBRID CPU–FPGA ARCHITECTURE

    by A. Maidanov, H. Canbolat, S. Atanov

    The paper presents the development and optimization of a hybrid hardware-software implementation of the post-quantum cryptographic algorithm Saber on an embedded CPU-FPGA platform. The main objective of the research is to enhance the performance, energy efficiency, and security of post-quantum key exchange schemes under limited computational resources while maintaining resistance to side-channel attacks.The proposed architecture integrates the computational capabilities of the ARM processor and the FPGA core, enabling efficient distribution of workloads between the processor and the hardware accelerator.The hardware part implements pipelined polynomial multiplication and SHA-3 hashing, while the software component manages data flow, synchronization, and integrity control. A fixed-latency communication interface between the CPU and FPGA ensures constant-time execution and stability against timing variations.Three implementations of the algorithm were compared: software, hardware, and hybrid. Experimental results demonstrated a 35-50% reduction in execution time without compromising cryptographic strength or increasing power consumption. A TVLA (Test Vector Leakage Assessment) analysis confirmed the absence of any statistical correlation between the energy profile and secret data, validating the system’s side-channel resistance.The proposed solution can be effectively applied to mobile robotic platforms, unmanned marine vehicles, industrial telemetry networks, and IoT systems requiring high-performance and quantum-resistant data protection.

  • Open Access

    Reconstruction of a Signal Sampled Slower than the Usual Nyquist Rate Using Complex Sampling

    by Hüseyin Canbolat

    In this paper, a modified sampling rate, which is smaller than the usual Nyquist rate, is determined for bandlimited signals using a method like single side band (SSB-like). SSB-like method is characterized using the base signal of single side band Amplitude Modulation (SSB-AM) instead of the actual AM signal. The sampling should be performed on the base SSB signal, which is a complex valued signal. That is, SSB-like sampling is performed on a complex valued signal with a corresponding Nyquist frequency which is half of the actual required sampling frequency. Also, a theoretical sampling method is proposed as a theorem that results in sampled values based on the mathematical properties of transformations. The proposed complex sampling method is novel in the context of sampling and valid for any band-limited real valued signal. The application of the method to real physical cases and a block diagram for complex sampling is under study.

  • İşitme Engelli Bireylerin Hareketlerini Sınıflandırmaya Yönelik Yapay Zeka Modelinin Geliştirilmesi

    by Ahmed Kasapbaşı, Hüseyin Canbolat

    İşaret Dili, işitme engelli bireyler için hayati bir iletişim aracıdır. Farklı ülkelerde kendi ihtiyaçlarına geliştirilmiş birçok işaret dili vardır. Bu çalışma, Türk İşaret Dili (TİD) jestlerini derin öğrenme teknikleriyle metne dönüştürmeyi amaçlamaktadır. Bu amaçla, arka planlar, aydınlatma koşulları ve işaret pozisyonları gibi çeşitli çevresel faktörler açısından çeşitlilik gösteren yeni bir veri kümesi oluşturulmuştur. Daha sonra, TİD alfabesini algılamak ve sınıflandırmak için Evrişimli Sinir Ağları (CNN'ler) kullanılmıştır. Ayrıca, geliştirilen modellerin performansını optimize etmek için çeşitli hiperparametreler araştırılmıştır. En iyi CNN mimarisi, beş evrişimli katmanı içerir ve Adam öğrenme hızı optimizasyon yöntemini kullanır; 80 epoch'tan sonra yaklaşık %98'lik bir doğruluk (başarı) elde edilmiştir. Sonuç olarak, zorlu bir veri kümesi üzerinde eğitilen önerilen modeller, işaret dili tanıma alanında önemli bir ilerleme temsil etmektedir.


  • Open Access

    Differential Diagnosis of Diabetic Foot with Deep Learning Methods

    by Maide Çakir Bayer, Hüseyin Canbolat, Gökalp Tulum

    Diabetic foot complications, caused by prolonged hyperglycemia, are a significant health concern among diabetes patients. Majority of patients develop diabetic foot complications, contributing significantly to diabetes-related hospital admissions. These complications include foot ulcers, infections, ischemia, Charcot foot, and neuropathy. They also increase the risk of amputation, affecting quality of life and putting strain on healthcare systems. At this stage, early diagnosis plays a vital role. The process of diagnosing involves not only identifying the presence or absence of a disease, but also categorizing the disease. In this study, we examine the use of deep learning methods in the diagnosis of diabetic foot conditions. It explores various aspects such as predictive modeling and image analysis. The study discusses the progression of model designs, data sources, and interpretability methodologies, with a focus on improving accuracy and early detection. Overall, the study provides a comprehensive analysis of the current state of deep learning in diabetic foot problems with highlighting advancements.

  • Open Access

    Mikro Şebekeler için Enerji Yönetimi ve Gerilim Kontrol Algoritmalarının Geliştirilmesi: Alçak Gerilim Seviyesinde Statik Voltaj Regülatörü Örneği

    by Enes Bektaş, Kamil Çağatay Bayindir, Alper Terci̇yanli, Adnan Tan, Hüseyin Canbolat, Hasan Yilmaz

    Son yıllarda dağıtık enerji üretiminin artması ile birlikte alçak gerilim (AG) tarafta, tüketici geriliminde yükselmeler meydana gelmektedir. Aynı zamanda ters güç akışı, şebeke dengesizliği gibi problemler ile birlikte gerilimdeki yükselme, enerji hatları üzerinde kayıplara yol açarak sistemi veriminin düşmesine neden olmaktadır. Bu makalede, AG mikro şebekeler için gerilim problemlerinin çözümüne yönelik, mikro şebeke önüne statik voltaj regülatörü (SVR) bağlanmıştır ve SVR gerilim kontrol algoritması geliştirilmiştir. Gerilim kontrol algoritmasının mikro şebeke için önerilen Kural Tabanlı ve Optimizasyon Tabanlı Enerji Yönetimi Sistemleri (EYS) ile birlikte uygulanmasına yönelik benzetim çalışması yapılmıştır. IEEE 13 bara test sistemi ve önerilen algoritmalar Simulink/MATLAB ortamında oluşturulmuştur. Optimizasyon tabanlı EYS sonuçları, Python programlama dili kullanılarak elde edilmiştir. Gerçek gerilim ve yük profili otomatik sayaç okuma sistemi (OSOS) verilerinden alınmıştır ve elde edilen sonuçlar SVR gerilim kontrol algoritmasının, mikro şebekelerinin gerilimini istenilen seviyeye indirilmesinde etkili bir şekilde kullanılabileceğini göstermiştir. Aynı zamanda Optimizasyon Tabanlı EYS ile mikro şebekelerin daha efektif bir şekilde yönetilebileceği benzetim çalışması sonuçları ile doğrulanmıştır.


  • Prediction of Turkish Sign Language Alphabets Utilizing Deep Learning Method

    by Ahmed Kasapbaşi, Hüseyin Canbolat

    Sign Language (SL) is an important tool for communicating the hearing-impaired persons. Therefore, many sign languages exist depending on the language of the country. This study proposes using Turkish sign language (TSL) alphabets to exploit the evolution of deep learning (DL) science to convert mimes of sign language to text. Therefore, a new dataset is created with novelty in the type and conditions. The created dataset includes different conditions in backgrounds, lighting, and positions. Then, this paper suggests Convolutional Neural Networks (CNNs) models detect and classify the TSL alphabets signs. Moreover, the research tries to apply different hyperparameters to the developed models to get the best result. The best model of CNN contains five convolutional layers and uses the Adam learning rate method to adjust the accuracy. This model gives 97.80% accuracy after 70 epochs. Consequently, the suggested models with the challenging dataset are an important point for developing the sign language recognition field.