Lightweight Vehicle Detection and Recognition Method Based on Improved YOLOv5 in SAR Images

Synthetic aperture radar (SAR) is very widely used in the military and civilian fields, but the target recognition and detection in the SAR images are relatively difficult. The existing deep learning methods for the SAR target detection and recognition has the problem of many model parameters and low efficiency, thus this paper has proposed a lightweight vehicle detection and recognition method based on the improved YOLOv5 in the SAR images. First, the lightweight processing of reducing the number of the channels is carried out, which can reduce many parameters and speed up the inference. Then, because the vehicles in the SAR image are relatively small and the detection head for detecting the large targets is redundant, the lightweight processing of removing the detection head is performed, which can shorten the running time of the proposed algorithm. Finally, the convolutional block attention module (CBAM) has been added to make the training set fit better when the network model parameters are limited, which can avoid the problem of the excessive decline in the detection and recognition performance after the network model is reduced in the parameters. The experimental results are shown to verify the correctness and the effectiveness of the proposed method.