SKSercan Külcü

Sercan Külcü

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Sercan Külcü has a total of 5 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Impact of temporal resampling interval on deep learning-based vessel trajectory prediction using AIS data

    by Sercan Külcü, Adil Kondiloğlu

    This study examines how temporal resampling intervals affect deep learning-based vessel trajectory prediction using Automatic Identification System (AIS) data. For this purpose, a public AIS dataset from the National Oceanic and Atmospheric Administration (NOAA) was filtered, validated, and transformed into trajectory datasets with four different time intervals: 1 min, 5 min, 10 min, and 20 min. Six learning-based models were tested under the same experimental conditions and one-step-ahead prediction setting. The evaluated models were Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Bidirectional Long Short-Term Memory (BiLSTM), Transformer, Transformer-LSTM, and Transformer-GRU. A constant speed-course baseline was also used as a reference method. The results show that prediction error increases as the resampling interval becomes longer. Shorter intervals provide more detailed information about vessel motion and improve continuity between consecutive trajectory points. This leads to lower training loss, Average Displacement Error (ADE), and Root Mean Square Error (RMSE) values. Transformer-GRU showed the strongest performance at the 20 min interval. In contrast, standalone recurrent models became more sensitive when the temporal resolution became coarser. The baseline method also remained competitive for relatively smooth and approximately linear vessel motion. These results indicate that temporal resampling should not be regarded solely as a preprocessing step, as the resulting temporal configuration influences model behavior, prediction error, and comparative performance. Although the proposed analysis was validated using AIS datasets collected on three representative dates, future studies should investigate a broader range of temporal and operational conditions to further assess the generalizability of the findings.

  • COMPARATIVE ANALYSIS OF TARGET LOCALIZATION ACCURACY IN RADAR PPI IMAGES USING GEOMETRIC ENCLOSURES

    by Sercan Külcü

    Maritime surveillance system requires sensor fusion of radar Plan Position Indicator (PPI) imagery with Automatic Identification System (AIS) data. However, systematic azimuth errors (0°–5°) and irregular echo shapes reduce ship localization accuracy significantly. This study compares three geometric enclosure methods (bounding boxes, convex hulls, and concave hulls) using the publicly available DLR Baltic Sea dataset (X-band, ~10 m resolution, 781 synchronized scans of 4 ships). A 4.5° azimuth offset compensation is applied to minimize target position deviation. Target positions are estimated from calculated bounding box centers and hull centroids from radar images. Validation against AIS ground truth data achieved the lowest average localization errors (58–70 m) for concave hull centroids. It demonstrates up to 12.7% improvement over bounding boxes solution and a 2.1% improvement over convex hulls solution. The results show the accuracy of the concave hulls method in target detection and in preserving echo concavities caused by sea clutter, and noise etc. The concave hull method offers a lightweight and robust solution for real-time Vessel Traffic Services (VTS) and autonomous surface vehicles (ASVs).

  • Open Access

    Overcoming resolution constraints in automated colony counting via a high-performance deep learning framework using SAHI

    by Sercan Külcü, Duygu Balpetek Külcü

    Lightweight models perform poorly in bacterial colony counting when high-resolution Petri dish images are downscaled to 640 × 640 pixels. This study addresses this issue using a tiled training and SAHI-based tiled inference pipeline. Full-resolution images are divided into overlapping 640 × 640 tiles (20% overlap), preserving native resolution and reducing colony density per tile. A public 24-class bacterial colony dataset was used to evaluate the proposed approach. The tiled strategy substantially increased the number of training samples while preserving local colony details. Three nano-scale models (YOLOv5n, YOLOv8n, YOLOv11n) were evaluated. The conventional resize method yielded only 44.9-66.3% mAP@0.5, whereas the tiled approach achieved 95.4-96.9% mAP@0.5 and 55.4-58.4% mAP@0.5:0.95. YOLOv11n with tiling achieved the highest mAP@0.5 while using only 2.6 million parameters. It reached 96.9% mAP@0.5 and 58.2% mAP@0.5:0.95, and the validation-set confusion matrix showed class-wise correct prediction rates above 95% in 23 of 24 classes. Whole-plate inference required less than 320 ms on a laptop RTX 3050 GPU. For small, densely packed objects such as bacterial colonies, tiled inference with moderate overlap dramatically outperforms both image resizing and recent architectural advances in YOLO. The proposed lightweight pipeline is suitable for routine laboratory deployment.

  • Bias Mitigation in Ensemble-Based Meat Freshness Classification Using Grad-CAM

    by Sercan Külcü, Duygu Balpetek Külcü

    Visual biases in deep learning models, such as focusing on packaging trays instead of meat texture, reduce the reliability of computer vision systems in food safety applications. This study proposes a Grad-CAM-guided bias mitigation framework for multiclass meat freshness classification that combines explainable AI with a lightweight hybrid ensemble design. A MiniCAM attention module is integrated into MobileNetV2 to redirect model focus toward meat-specific visual cues, and its features are fused with complementary embeddings extracted from Xception. The final decision is obtained by combining the predictions of MobileNetV2 with classical classifiers (SVM and XGBoost) using test-time augmentation and grid-optimized weighted ensembling. The proposed framework achieves 99.78% accuracy on the held-out test set and 99.66% ± 0.23 average accuracy under 5-fold cross-validation, while maintaining real-time efficiency (4.3M parameters, 16.5 MB model size, and 825.1 FPS on a single GPU), and effectively suppresses non-informative background elements (e.g., packaging trays) as confirmed by Grad-CAM visualizations. These results demonstrate that integrating explainable bias mitigation with lightweight ensemble learning enables reliable and deployable meat freshness assessment for real-world food safety inspection.

  • Open Access

    Radar image-based object detection and tracking for autonomous surface vehicles

    by Sercan Külcü

    This study presents a modular and lightweight framework for radar image-based object detection and tracking, specifically designed for autonomous surface vehicles (ASVs). The detection module uses HSV (Hue, Saturation, Value) color-space segmentation to identify navigational targets, particularly yellow-encoded objects, and the own ship indicator, typically displayed in white on commercial marine radar screens. To enhance stability under varying sea conditions, the system applies Gaussian smoothing, morphological operations, and area-based filtering during preprocessing. The dynamic multi-target tracking module assigns persistent object identifiers (ID) using a Euclidean distance-based association scheme enhanced with velocity and direction-aware ID recovery. This approach reduces ID fragmentation caused by occlusions or sudden maneuvers. Furthermore, the framework estimates relative speed of each target, enabling inference of behaviors such as approaching or moving away from the own ship. The proposed detection method, evaluated on 7,200 radar frames, achieved a true positive rate of 98.7%, a 4.4% improving over the area-based baseline and a 59% reduction in the false positive rate. This demonstrates robust target discrimination in noisy and complex conditions. The proposed tracking system achieved significant gains in tracking continuity and identification stability, reducing ID switches by 75% and decreasing the total number of generated IDs by 58% compared to positional-only baseline. The average processing time of 240 ± 17 milliseconds per frame validates the framework’s suitability for real-time embedded deployment in dynamic maritime environments.


  • Open Access

    High-Precision Marine Radar Object Detection Using Tiled Training and SAHI Enhanced YOLOv11-OBB

    by Sercan Külcü

    Reliable object detection in marine radar imagery is critical for maritime situational awareness, collision avoidance, and autonomous navigation. However, it remains challenging due to sea clutter, small targets, and interference from fixed navigational aids. This study proposes a high-precision detection pipeline that integrates tiled training, Sliced Aided Hyper Inference (SAHI), and an oriented bounding box (OBB) variant of the lightweight YOLOv11 architecture. The proposed approach effectively addresses scale variability in Plan Position Indicator (PPI) radar images. Experiments were conducted on the real-world DAAN dataset provided by the German Aerospace Center (DLR). The dataset consists of 760 full-resolution radar frames containing multiple moving vessels, dynamic own-ship, and clutter sources. A semi-automatic contour-based annotation pipeline was developed to generate multi-format labels, including axis-aligned bounding boxes, oriented bounding boxes (OBBs), and instance segmentation masks, directly from radar echo characteristics. The results demonstrate that the tiled YOLOv11n-OBB model with SAHI achieves an mAP@0.5 exceeding 0.95, with a mean center localization error below 10 pixels. The proposed method shows better performance on small targets compared to standard full-image baselines and other YOLOv11 variants. Moreover, the lightweight models enable near real-time inference at 4–6 FPS on edge hardware. These findings indicate that OBBs and scale-aware strategies enhance detection precision in complex marine radar environments, providing practical advantages for tracking and navigation tasks.


  • Open Access

    Integration of Steerable Smart Antennas to IETF 6TiSCH Protocol for High Reliability Wireless IoT Networks

    by Sercan Kulcu, Sedat Gormus, Yichao Jin

    Steerable directional antennas are increasingly utilised to improve the overall performance of the traditional wireless sensor networks. Steerable directional antenna based networking solutions increase the network capacity by providing a longer range of transmission and reduced interference as compared to networking solutions with omni-directional antennas. However, the use of smart antennas requires complex algorithms and such algorithms may not be easily leveraged in low power Internet of Things (IoT) networks. This study presents mechanisms for integrating low complexity smart antenna solutions into IETF 6TiSCH protocol with the aim of creating scalable and reliable industrial IoT networks. The solution defines extensions to MAC layer and scheduling mechanisms of IETF 6TiSCH protocol to enable its seamless integration with low complexity steerable smart antennas. The results of this study show that smart antenna enabled 6TiSCH protocol stack outperforms the legacy 6TiSCH stack in terms of data delivery performance especially in high density scenarios.


  • Open Access

    Integration of Steerable Smart Antennas to IETF 6TiSCH Protocol for High Reliability Wireless IoT Networks

    by Sercan Kulcu, Sedat Gormus, Yichao Jin

    Steerable directional antennas are increasingly utilised to improve the overall performance of the traditional wireless sensor networks. Steerable directional antenna based networking solutions increase the network capacity by providing a longer range of transmission and reduced interference as compared to networking solutions with omni-directional antennas. However, the use of smart antennas requires complex algorithms and such algorithms may not be easily leveraged in low power Internet of Things (IoT) networks. This study presents mechanisms for integrating low complexity smart antenna solutions into IETF 6TiSCH protocol with the aim of creating scalable and reliable industrial IoT networks. The solution defines extensions to MAC layer and scheduling mechanisms of IETF 6TiSCH protocol to enable its seamless integration with low complexity steerable smart antennas. The results of this study show that smart antenna enabled 6TiSCH protocol stack outperforms the legacy 6TiSCH stack in terms of data delivery performance especially in high density scenarios.