MITD-YOLO: Maritime Infrared Target Detection Method Based on YOLOv8n

Objective The complex backgrounds, significant variations in target sizes, and interference from ocean wave clutter in maritime infrared images often lead to missed and false detections. To enhance the accuracy of target detection in infrared images, a novel maritime infrared target detection method based on YOLOv8n, named MITD-YOLO (Maritime Infrared Target Detection-YOLO) was proposed. Methods The method integrates a Diversified Branch Block (DBB) and Enhanced Multi-Scale Convolution (EMSConv) modules. It employs a triple attention mechanism to facilitate feature interaction in spatial and channel dimensions, thereby strengthening the extraction of critical features. Additionally, the loss function of the original model is improved by incorporating Powerful-IoUv2 (PIoUv2), which enhances detection accuracy and robustness. Results Experimental results demonstrated that MITD-YOLO achieved a precision of 91.7% in maritime infrared target detection, representing a 2.3% improvement over the YOLOv8n model. The recall rate reached 82.4%, with an increase of 1.7%, and the average precision reaches 88.9%, marking an improvement of 20.2%. Furthermore, the method achieved a frame rate of 132.8 FPS. Conclusion The proposed method effectively improves the efficiency of maritime infrared target detection.