视觉传感信息耦合零件语义关系与多任务学习的输电线路螺栓缺陷检测方法OA

Targeting at the problems of low efficiency of manual inspection in the process of transmission line bolt defects,and the limi-tation of existing visual detection methods for single-task recognition which leads to the neglect of component semantic correlation,an in-telligent bolt defect detection method integrating part semantic relationship and multi-task learning is proposed.First,a semantic rela-tionship model of bolts is constructed.By defining geometric and structural attributes such as exposed thread length,nut gap,and incli-nation angle,the mapping rules between these attributes and defect types such as loosening,corrosion,and cracks are established.Then,a multi-task deep learning network is designed.Through shared backbone feature extraction,tasks including defect classification,seman-tic attribute regression,and component localization for transmission line bolt detection are simultaneously implemented.In addition,a se-mantic consistency constraint loss is introduced to enhance the interpretability and generalization ability of the model.Experiments on the self-built transmission line bolt dataset show that the average precision of the proposed method in defect detection reaches 85.6%,which is more than 5%higher than that of the single-task baseline model,the regression error of semantic attributes is reduced by more than 4.2%,and the mean intersection over union(mIoU)of component segmentation reaches 84.1%.The method improves the accuracy and robustness of bolt defect detection.Based on the detection results of semantic attributes,it can provide an interpretable basis for the operation and maintenance decision-making of transmission lines,thus offering a new visual detection solution and idea for the intelligent inspection of transmission lines.