In order to solve the challenges faced by target detection algorithms of UAVs in transmission line insulator inspection, such as high model complexity, insufficient accuracy in detecting defects of small targets, and easy feature loss during up and down sampling, this paper proposes a lightweight improved RT-DETR insulator defect detection algorithm based on lightweight improvement (SDH-DETR). Firstly, RT-DETR is used as the baseline algorithm to reduce the optimisation difficulty and improve the robustness; secondly, lightweight StarNet is used as the backbone network to improve the feature extraction capability while significantly reducing the model complexity; next, the DySample dynamic upsampling module is introduced to efficiently reduce the detail loss and image distortion by the adaptive upsampling method based on the sampling points. Finally, the Harr wavelet transform downsampling module (HWD) is used to achieve efficient fusion of low-frequency and high-frequency information, suppressing complex background interference and enhancing the detection of small targets. The validation experiments on the complex background dataset show that the average accuracy of SDH-DETR reaches 98.5%, which is 0.9% higher than the baseline algorithm, the number of parameters and computation are reduced by 43% and 46.1%, respectively, and the detection speed reaches 78.6 fps. This indicates that the algorithm achieves a lightweight design while ensuring high accuracy, and meets the practical demands for efficiency and performance in transmission line inspection.