The diversity of transmission line insulator defect characteristics largely affects their identification.The present work hence made a preliminary attempt to introduce XGBoost algorithm into the identification of transmission line insula-tor defects.The main efforts entailed first the enhancement processing of original insulator images via grayscale transfor-mation in spatial domain method and template convolution-based image smoothing technique,second the establishment of a strong classifier by integrating multiple weak classifiers (CART regression tree)and an XGBoost objective function con-taining loss function and normalization term,and third the optimum solution through Taylor expansion and second-order derivative information.The designed method achieved in the tentative test effective identification for most of the defects and false rate within 1.0％ in a handful of cases,demonstrating potential identification utility.
维普
万方数据