Abstract: Aiming at the current problem of feature redundancy and low detection accuracy in the detection process caused by the small number of defective samples and complex background of transmission line insulators, this paper proposes a target detection model based on the sparse reconstruction dual attention (SRDA) mechanism. Firstly, to mitigate the influence of redundant deep features, it employs a sparse reconstruction mechanism to filter the deep feature layer of the model. Secondly, to enrich the model's capability in delineating target regions across various contexts, the paper introduces a positional attention mechanism capturing contextual cues from the shallow feature target region. Thirdly, by integrating a channel attention mechanism to augment the feature representation of specific semantic categories within the deep feature layer, the semantic portrayal of defective targets is enhanced. Finally, the research conducts defect detection experiments using UAV-captured images of insulators on the transmission lines. The results demonstrate the model's efficacy in discerning accurate defect features, thus improving the detection accuracy of insulator defects, surpassing performance benchmarks set by other models.