Research on military target detection method based on YOLO method
Huibai Wang, Han Ji · 2023
In the field of military target recognition, in order to solve the problems of easy loss of small target detection results and slow detection speed, this paper proposes a target recognition detection algorithm YOLO-G2S, which reduces the parameters and reduces the model size by sacrificing a little accuracy. In the original model of YOLO v5, the 2nd, 3rd and 4th C3 modules in the backbone were replaced by C3G2 modules, thus reducing the model size and improving the detection speed. Then the accuracy is improved by changing the RELU activation function in C3G2 module. The experimental results show that the average accuracy of YOLO-G2S reaches 95%, the number of parameters is reduced by 7.3%, and the amount of calculation is reduced by 17.8%. Therefore, the improved algorithm in this paper is more suitable for deployment in the field of military target recognition.