Infrared dim small target detection algorithm based on improved YOLOX

Abstract Detection of infrared dim small targets in complex contexts has always been a difficult area of infrared target detection. Previous studies have shown that the traditional target detection methods are mainly based on the gradient, gray scale, contrast and other features of small targets in the infrared image. These features are mostly based on manual selection, and few are optimized due to the nature of the application field. By comparing and analyzing the performance of the Yolo series network, this paper adopts the YOLOX network framework and combines the cross-patch interactive attention mechanism attention module to improve the feature extraction ability of the backbone network. The experimental results show that the detection rate of the model can reach 95.58% and the average detection accuracy is 94.79%, which is 0.5% and 1.47% better than YOLOX respectively.