The wire ropes of the hoisting machine are used for gate lifing in hydropower stations,which is crucial for the safe and stable production of hydropower.However,traditional manual inspection methods have issues such as low efficiency and poor accuracy.Utilizing wire rope monitoring images for defect localization can not only significantly improve inspection eficiency but also achieve highly accurate defect localization.This paper proposes a U-Net-based method that extracts multi-scale image features through an encoder and then restores these features into defect localization labels using a decoder,thereby realizing defect localization in wire ropes.The experimental results show that the proposed method significantly outperforms traditional fully convolutional networks and surpasses the comparison algorithms by 0.29,0.23,and 0.0047 in terms of the Dice coefficient,IoU,and Hausdorff distance,respectively, enabling more accurate wire rope defect localization.