Abstract: In the transmission line UAV inspection task, to address the problems of low detection accuracy of targets to be detected in deep learning-based aerial images and the model being too large and complicated to be deployed to mobile devices such as UAVs, a method of improving YOLOv7-tiny as the base network to achieve improved detection accuracy and light_weight of the model is proposed. Firstly, this paper designs an Interlace Partial Convolution (IPConv) and uses it to construct IP1-ELAN and IP2-ELAN modules as the feature extraction module of the network, so that it can effectively alleviate the problem of channel redundancy in the model and substantially reduce the number of parameters and floating points in the model; secondly, in the last layer of backbone Secondly, Efficient Multi-Scale Attention (EMSA) is integrated in the previous layer of the network to achieve cross-channel interaction and enhance the feature extraction capability of the target region; finally, SPPFCSPC (Spatial Pyramid Pooling Faster, Cross Stage Partial Channel) module to further enhance the feature extraction capability and improve the model detection performance. Through experimental verification, the number of model parameters and floating points in this method's transmission line inspection dataset are only 3.79M and 8.4G, respectively, and the detection accuracy is 85.8%. The comprehensive performance is better than that of the current commonly used detection algorithms, and can basically meet the deployment to the UAV end for detection tasks.