Due to the characteristics of metal surface defect images,effective an d accurate segmentation is a major challenge in image processing tasks.In order to obtain defect type,size and location information,this paper proposes a met al defect image segmentation network that incorporates attention mechanism.The network is divided into two paths.The semantic information path is mainly compo sed of a convolutional network composed of residual blocks to obtain a feature m ap.During the sampling process,the attention mechanism is integrated step by s tep to enhance the contrast between the feature and the background.The side roa d path design attention mechanism module obtains the weight map of the location information,and then combines the feature map of the same size with the weight map,and combines multi-scale features through the spatial pyramid.Experimenta l results show that the algorithm can improve the segmentation accuracy of metal surface defect images.