A lightweight detection method for infrared ship targets based on improved YOLOX

Aiming at the problem of low detection accuracy and poor real-time performance in infrared ship target detection, a lightweight detection method based on improved YOLOX is proposed, which can effectively improve detection real-time performance while ensuring detection accuracy. The algorithm first adopts an anchor free box design, then improves the backbone and neck of the network, introduces task alignment learning methods, assigns labels, and finally strengthens the detection head in a targeted manner. Through experimental verification, the mAP of this method can reach 50.1%, and the FPS can reach 78.3%, with better accuracy and real-time performance than other similar algorithms.