GT-YOLO is an improved detection model based on YOLOv5s for nearshore infrared ship detection. The model addresses challenges in detecting small and dense targets in coastal environments by introducing a feature fusion module with a fusion attention mechanism, SPD-Conv to enhance small target detection, and Soft-NMS to handle dense occlusion. These improvements result in increased mAP values of 1%, 5.7%, and 5% on an infrared ship dataset. The model achieves better detection performance in complex sea conditions, with enhanced accuracy and robustness in detecting small targets and dense scenes. The algorithm is evaluated using metrics such as precision, recall, mAP, GFLOPS, parameters, and FPS, demonstrating significant improvements over existing methods. GT-YOLO outperforms other algorithms like YOLOv5m, YOLOv8s, SSD, CenterNet, and EfficientDet-D0 in detection accuracy and efficiency. The study highlights the effectiveness of the proposed algorithm in nearshore ship detection, with future work focusing on further optimizing the model to reduce computational load and improve real-time performance.GT-YOLO is an improved detection model based on YOLOv5s for nearshore infrared ship detection. The model addresses challenges in detecting small and dense targets in coastal environments by introducing a feature fusion module with a fusion attention mechanism, SPD-Conv to enhance small target detection, and Soft-NMS to handle dense occlusion. These improvements result in increased mAP values of 1%, 5.7%, and 5% on an infrared ship dataset. The model achieves better detection performance in complex sea conditions, with enhanced accuracy and robustness in detecting small targets and dense scenes. The algorithm is evaluated using metrics such as precision, recall, mAP, GFLOPS, parameters, and FPS, demonstrating significant improvements over existing methods. GT-YOLO outperforms other algorithms like YOLOv5m, YOLOv8s, SSD, CenterNet, and EfficientDet-D0 in detection accuracy and efficiency. The study highlights the effectiveness of the proposed algorithm in nearshore ship detection, with future work focusing on further optimizing the model to reduce computational load and improve real-time performance.