基于稀疏重构注意力机制的绝缘子缺陷检测方法

Insulator Defect Detection Based on Sparse Reconstruction Dual Attention Mechanism

针对当前输电线路绝缘子缺陷样本数量少、缺陷目标背景复杂干扰导致检测过程中出现的特征冗余以及检测精度低等问题,提出基于稀疏重构注意力(sparse reconstruction dual attention,SRDA)机制的目标检测模型。首先,为了降低深层特征冗余对模型的影响,采用稀疏重构机制对模型的深层特征层进行筛选和过滤;其次,为了增强模型对不同背景下目标区域的表达能力,提出位置注意力机制来捕获浅层特征目标区域的上下文信息,并引入通道注意力机制在深层特征层上加强对特定类别语义的特征表示,增强缺陷目标的语义信息;最后,通过对无人机拍摄采集的输电线路绝缘子图像进行缺陷检测实验,证明该模型能够获取精确的缺陷特征,提高绝缘子缺陷检测精度,与其他模型相比,该模型具有一定的优越性。

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.