ReNet: Ground Rotating Target Detection Method Based on Anchor-Free Frame

Ground infrared target detection is crucial in the fields of high-altitude reconnaissance, intelligent perception, and ground strike, where the acquired ground infrared targets often appear in the form of irregular angles, resulting in low detection accuracy, ease of misdetection, and other problems. Therefore, this paper proposes an anchor-free-based ground rotating target detection method. Based on the anchor-free target detection model, a backbone network based on atrous convolution is constructed, which enhances the perception range and feature extraction ability of the model for ground rotating targets. After feature extraction based on void convolution, the attention dimension of the extracted feature is increased through external attention, and the extraction of higher-resolution features of the target is realized. The ground rotating target detection model achieved 90.6% detection accuracy on the HIT-UAV dataset, which optimized the detection performance of the anchor-free target detection model for ground rotating targets.