Infrared Ship Target Detection Algorithm Based on Improved YOLOV8

Infrared ship detection in complex sea conditions faces challenges of multi-scale targets, blurred edges, and strict real-time demands. This paper proposes an enhanced algorithm based on improved YOLOv8. We introduce an Edge-enhanced Visual Center (EVC) module during feature extraction to boost contour feature representation through multi-scale fusion and edge gradient reinforcement. The backbone is replaced with efficient HGNetV2, optimizing feature flow and reducing redundancy. Experiments on public datasets show the improved model achieves 93.7% mAP@0.5 while maintaining real-time operation, outperforming the baseline YOLOv8 by +4.8% mAP and +3.6% recall. Ablation confirms EVC enhances small-target accuracy, while HGNetV2 increases inference speed by 4.2%. The algorithm demonstrates superior accuracy-speed balance.