An improved YOLOv3 infrared small target detection algorithm

Infrared small target detection technology plays an important role in military early warning and precise guidance. In this paper, an improved YOLOv3 infrared small target detection algorithm is proposed, which mainly includes three parts: feature extraction, feature fusion and regression prediction. In the feature extraction stage, the original 32 -magnification downsampling detection head is replaced by a 4-magnification downsampling detection head to adapt to the sparse characteristics of small targets. The fused multiple feature maps are respectively focused by infrared IAM to achieve the feature focus of the region of interest. From the experimental comparison results, it can be seen that the proposed method has a great improvement compared with SSD, YOLOv3 and YOLOv4.