The detection of infrared multi-target Image and video in complex background is the hotspot and difficulty of target detection. In order to detect infrared target in complex background more accurately, the algorithm of YOLOv3 is improved. Firstly, by increasing the feature scale on the basis of the original algorithm, the recognition accuracy of remote and complex background image is improved, and the BN network layer and convolution neural network are combined. The final detection results are obtained by layer fusion calculation. The analysis and comparison between the original algorithm and the improved algorithm show that the improved algorithm can better the average recognition accuracy from 64% to 88%, and the mAP from 51. 73 to 59. 28, which is verified that the improved YOLOv3 algorithm has better performance and more obvious advantages in infrared target detection.