Publications

This study combines radar-specific preprocessing with YOLO-based deep learning to improve the detection of small and low-contrast maritime targets in PPI images. The proposed preprocessing, P2 feature layer, and area-aware loss substantially improve detection performance on both simulated radar data and the real-world DAAN dataset.

How to cite: Külcü, S. (2026). Enhancing Radar PPI-Based Maritime Object Detection Using Hybrid Preprocessing and Deep Learning Approach. IEEE Access, 14, 111715–111730.

Paper · Presentation

This study addresses the loss of small bacterial colonies caused by resizing high-resolution Petri dish images by combining tiled training with SAHI-based inference. The approach raises mAP@0.5 to as high as 96.9% with lightweight YOLO models while remaining fast enough for routine laboratory use.

How to cite: Külcü, S., & Balpetek Külcü, D. (2026). Overcoming resolution constraints in automated colony counting via a high-performance deep learning framework using SAHI. Scientific Reports, 16, 24516.

Paper · Presentation

This study compares bounding boxes, convex hulls, and concave hulls for estimating vessel positions from marine radar PPI imagery using AIS data as ground truth. Concave-hull centroids achieve the lowest average localization error and provide a lightweight solution for real-time maritime surveillance and autonomous surface vehicles.

How to cite: Külcü, S. (2026). Comparative Analysis of Target Localization Accuracy in Radar PPI Images Using Geometric Enclosures. Konya Journal of Engineering Sciences, 14(2), 901–916.

Paper · Presentation

This study introduces a Grad-CAM-guided framework that reduces visual bias in meat freshness classification by directing model attention away from packaging and toward meaningful meat characteristics. The lightweight ensemble reaches 99.78% test accuracy while maintaining real-time performance and strong explainability.

How to cite: Külcü, S., & Balpetek Külcü, D. (2026). Bias Mitigation in Ensemble-Based Meat Freshness Classification Using Grad-CAM. Balkan Journal of Electrical and Computer Engineering, 14, 74–82.

Paper · Presentation

This study integrates tiled training, SAHI inference, and YOLOv11 oriented bounding boxes for detecting vessels in challenging marine radar imagery. The method exceeds 0.95 mAP@0.5 on the real-world DAAN dataset and provides accurate target orientation and localization with near real-time performance on edge hardware.

How to cite: Külcü, S. (2026). High-Precision Marine Radar Object Detection Using Tiled Training and SAHI Enhanced YOLOv11-OBB. Sensors, 26(3), 942.

Paper · Presentation

This study proposes a lightweight radar-image framework that combines HSV-based target detection with multi-target tracking and velocity-aware ID recovery for autonomous surface vehicles. Evaluation on 7,200 radar frames shows a 98.7% true-positive rate, fewer false positives, and substantially improved tracking continuity.

How to cite: Külcü, S. (2026). Radar image-based object detection and tracking for autonomous surface vehicles. Turkish Journal of Maritime and Marine Sciences, 12(1), 35–47.

Paper · Presentation

This study presents mechanisms for integrating low-complexity steerable smart antennas into the IETF 6TiSCH protocol stack. The proposed scheduling, neighbor-discovery, and direction-selection mechanisms improve data delivery and scalability, particularly in dense industrial IoT networks.

How to cite: Külcü, S., Görmüş, S., & Jin, Y. (2021). Integration of Steerable Smart Antennas to IETF 6TiSCH Protocol for High Reliability Wireless IoT Networks. IEEE Access, 9, 147780–147790.

Paper · Presentation

This study investigates the integration of low-complexity directional antennas with IETF 6TiSCH to extend scheduling from the time and frequency domains into the spatial domain. The proposed directional-antenna-aware resource allocation improves spatial reuse and network performance in dense low-power IoT environments.

How to cite: Görmüş, S., & Külcü, S. (2019). Enabling space time division multiple access in IETF 6TiSCH protocol. Turkish Journal of Electrical Engineering and Computer Sciences, 27(6), 4151–4166.

Paper · Presentation

This review examines the use of computer vision for food quality assessment, including defect detection, contamination analysis, classification, and automated inspection. It highlights the advantages of rapid, objective, consistent, and non-destructive image-based evaluation for modern food-processing systems.

How to cite: Balpetek Kulcu, D., & Kulcu, S. (2018). Computer Vision Technology on Food Science. Karaelmas Science and Engineering Journal, 8(1), 403–409.

Paper · Presentation

This study compares YOLOv11-seg and YOLOv12-seg for cell-nucleus instance segmentation on the heterogeneous Data Science Bowl 2018 dataset. The results show that YOLOv11-seg provides stronger overall segmentation accuracy, boundary quality, recall, and inference performance.

How to cite: Külcü, S. (2025). Nucleus Segmentation in Heterogeneous Microscopy Images on the Data Science Bowl 2018 Dataset. In Proceedings of the 3rd International Conference on Recent and Innovative Results in Engineering and Technology (ICRIRET 2025), pp. 105–114. All Sciences Academy.

Paper · Presentation

This study combines EfficientNet-B0, DenseNet201, and ResNet50 with an MLP meta-learner for four-class brain-tumor classification on the BRISC MRI dataset. With a dedicated preprocessing pipeline and mixed-precision training, the ensemble achieves a weighted F1-score of 0.998.

How to cite: Külcü, S. (2025). Meta-Learner Ensemble Model for Brain Tumor Classification. In Proceedings of the 7th International Conference on Engineering and Applied Natural Sciences (ICEANS 2025), pp. 127–133. All Sciences Academy.

Paper · Presentation

This study benchmarks Transformer-based RT-DETR-L against YOLOv12n and YOLOv8n for underwater object detection on 9,200 sonar images from the UATD dataset. RT-DETR-L achieves the highest detection accuracy, while the YOLO models provide substantially faster inference and lower computational cost.

How to cite: Külcü, S. (2025). Transformer vs. CNN: Benchmarking RT-DETR and YOLO Models on the UATD Dataset. In 17th International Academic Studies Conference (UBCAK) Full Text Book, pp. 328–337. Asos Publishing.

Paper · Presentation

This study evaluates SGDM, ADAM, and RMSprop optimizers with ShuffleNet for classifying hazelnuts into good, bad, and inner categories using 15,770 images. ADAM gives the best overall result, reaching 99.96% test accuracy and an F1-score of 0.9998.

How to cite: Tomak, Ö., Dikbaş, M. C., & Külcü, S. (2025). Selection of the Most Suitable Parameters for Classification of Hazelnut Fruit Using ShuffleNet Deep Learning Algorithm. In 11th International Azerbaijan Congress on Life, Engineering, Mathematical, and Applied Sciences, Congress Proceedings Book, Vol. I, pp. 64–73. BZT Turan Publishing House.

Paper · Presentation

This study compares ResNet-50 and ShuffleNet on a 15,770-image hazelnut dataset containing good, bad, and inner-hazelnut classes. ResNet-50 gives slightly higher accuracy, while ShuffleNet is recommended for real-time use because of its lower computational requirements and faster operation.

How to cite: Tomak, Ö., Dikbaş, M. C., & Külcü, S. (2025). Comparison of the Performance of ResNet-50 and ShuffleNet Deep Learning Algorithms in Classification of Hazelnut Fruit. In 11th International Azerbaijan Congress on Life, Engineering, Mathematical, and Applied Sciences, Congress Proceedings Book, Vol. I, pp. 55–63. BZT Turan Publishing House.

Paper · Presentation

This study proposes a new 6TiSCH message mechanism that allows nodes equipped with smart antennas to discover neighbors before completing network synchronization. Experimental results show that the proposed mechanism reduces synchronization time compared with conventional operation.

How to cite: Külcü, S., & Görmüş, S. (2022). Improving Synchronization Time in 6TiSCH Networks with Smart Antennas. In 2022 30th Signal Processing and Communications Applications Conference (SIU), pp. 1–4. IEEE.

Paper · Presentation

This study introduces a new RPL parent-selection objective function for IETF 6TiSCH networks using neighbor count and routed traffic as decision factors. The proposed approach reduces parent switching and improves message-delivery performance in the evaluated network scenarios.

How to cite: Aydın, B., Görmüş, S., Aydın, H., & Külcü, S. (2022). A New Routing Objective Function for IETF 6TiSCH Protocol. In 2022 30th Signal Processing and Communications Applications Conference (SIU), pp. 1–4. IEEE.

Paper · Presentation

This survey reviews systems and technologies that combine Semantic Web methods, Big Data platforms, and analytics for social-network analysis. It summarizes the state of the art and identifies research directions for processing and integrating large, heterogeneous social-network datasets.

How to cite: Kulcu, S., Dogdu, E., & Ozbayoglu, A. M. (2016). A Survey on Semantic Web and Big Data Technologies for Social Network Analysis. In 2016 IEEE International Conference on Big Data (Big Data), pp. 1768–1777. IEEE.

Paper · Presentation

This work presents a scalable framework that classifies Turkish tweets by sentiment and links them to related news items using machine-learning and NLP techniques. Experiments on 6,000 manually labeled tweets show strong performance for Naive Bayes-based methods in Turkish sentiment classification.

How to cite: Kulcu, S., & Dogdu, E. (2016). A Scalable Approach for Sentiment Analysis of Turkish Tweets and Linking Tweets to News. In 2016 IEEE Tenth International Conference on Semantic Computing (ICSC), pp. 471–476. IEEE.

Paper · Presentation

This PhD thesis investigates the integration of smart and directional antennas with the IETF 6TiSCH protocol stack to improve resource utilization in low-power IoT networks. It proposes neighbor discovery mechanisms, antenna direction selection methods, and hybrid distributed/autonomous and centralized/autonomous scheduling algorithms that exploit spatial reuse in addition to time and frequency resources.

Original title: IoT Ağlar İçin Akıllı Anten Tabanlı Çizelgeleme Algoritmalarının İncelenmesi

How to cite: Külcü, S. (2022). Analysis of Smart Antenna Based Scheduling Algorithms for IoT Networks [PhD thesis, Karadeniz Technical University].

Thesis · Presentation EN · Presentation TR