SwinT-YOLOv5s: Improved YOLOv5s for Vehicle-mounted Infrared Target Detection

Abstract

Infrared vehicle-mounted target detection is an important research direction in assisted driving, but also a very challenging topic. Existing infrared target detection methods often have problems such as high missed detection rate and false alarm in complex background, small target size and occlusion scene. A SwinT-YOLOv5s algorithm is proposed by the fusion of attention mechanism and convolutional network. Based on YOLOv5s algorithm, a detection layer is added to enhance the detection ability of small target objects. The CBAM modules are inserted into the backbone network to make the model pay more attention to the useful information and resist the interference of redundant information, so as to improve the detection ability in dense scenes. In addition, the Swin Transfomer encoders are used to replace some part of C3 modules to improve the model's ability of mining potential feature details and further improve the detection accuracy of the model. Experimental results show that the improved algorithm increases the average precision (IOU=0.5) and precision rate by 5.60% and 4.20% compared with the original YOLOv5s model, and has good generalization ability in remote small target and overlapping target scenarios.