Aiming at the problems of low accuracy and poor effect of UAV intelligent power inspection caused by the complex background and dense small targets of power equipment, an improved target detection algorithm of YOLOv5. Firstly, a detection layer is added to the original model to re-obtain the anchor frame so as to better learn the multi-level features of dense small targets and improve the ability of the model to deal with complex power scenarios. Secondly, the feature fusion module PANet structure of the model is improved, and the features of different scales are fused by jumping connection to enhance the dissemination and reuse of information. Finally, combined with the collaborative attention module, the backbone network is designed to focus on the target characteristics and enhance the visibility of dense target areas in complex backgrounds. The experimental results show that the average accuracy of the proposed algorithm (IoU=0.5) reaches 97.1%, which is 5.6% higher than the original network detection performance, and effectively improves the false detection and missed detection of small targets in complex background.