GCM-YOLO: A Multi-Scale Multi-View Infrared Ship Target Detection Model Based on the Improved YOLOv8

The detection of infrared ship target constitutes a cornerstone in areas including infrared guidance, satellite detection, and maritime warfare. Existing infrared ship target detection models can achieve good detection accuracy in simple, general scenes. However, under complex sea-sky scenarios, adverse weather conditions, and other challenging environments, their performance, including detection rate and false alarm rate, is significantly affected. This is particularly evident when facing problems such as drastic changes in target scale and substantial differences in target imaging from different viewpoints, where the performance of target detection models needs improvement. To address these issues, this study develops an enhanced ship detection model based on YOLOv8, named GCM-YOLO, for multi-scale, multi-view infrared ship target detection. The model introduces the GhostConv module to replace traditional convolution modules, improving inference speed. It also integrates the CBAM module and replaces SPPF with SPPFCSPC, enhancing the model's ability to detect complex features and improving multi-scale information fusion capability. Additionally, GCM-YOLO adds a small target detection head to improve the detection accuracy of small-scale ships. Our experiment demonstrate that GCM-YOLO outperforms traditional methods on a self-built infrared ship target dataset, especially demonstrating better accuracy and robustness in multi-scale, multi-view infrared ship target detection. This research provides an efficient and accurate solution for multi-scale, multi-view infrared ship target detection.