Aiming at the problems of large model size and poor real-time performance in the detection of transmission line components and defects

as well as the problems of frequent false detection and missed detection due to complex background and imbalanced samples

a light weight detection method for typical transmission line components and defects based on improved YOLOX was proposed. Firstly

Ghostnet was introduced to replace the original backbone network to achieve model lightweighting. Secondly

in order to solve the problem of frequent missed detection and false detection caused by the complex background of the transmission line image

the coordinate attention module was embedded into both of the feature fusion network and CSPLayer structure to enhance the focus on the target area in complex background. Finally

to address the problems of the imbalance samples of transmission line components and defects and the great differences in detection difficulty of different targets

the Varifocal loss function was introduced to reduce the impact caused by the imbalance of positive and negative samples

and improve the training intensity of difficult-to-classify targets to reduce false detection and missed detection. The experimental result shows that the average accuracy of the proposed method in the detection of transmission line components and defects is increased by 2.67%

the model size is reduced by 33.01%

and the reasoning speed is 19.61 ms/piece

which can satisfy the requirements of UAV lightweight and high-precision power real-time inspection.