In order to solve the problems that the size of different types of defects in magnetic tile images affects the segmentation accuracy, the detection model parameters are large and difficult to deploy in practical applications, and the image pixel distribution is uneven, a magnetic tile surface defect segmentation algorithm based on improved Deeplabv3+ was proposed. Firstly, in terms of structure, the lightweight MobilNetv3 network was used as the backbone network to replace the Xception network of the original model, so that the parameters and computational cost of the model were kept small to improve the detection speed. Secondly, the ECANet attention mechanism was introduced to improve the feature expression ability and generalization ability of the model. Finally, the loss function combining Dice Loss and Focal Loss were used to effectively alleviate the influence of sample pixel distribution imbalance on model training. The experimental results show that the average intersection union ratio of the proposed algorithm on the magnetic tile surface defects dataset was 68.25%, the average pixel accuracy was 82.80%, and the accuracy was 79.80%, compared with the original Deeplabv3+ algorithm, the average intersection union ratio was increased by 8.62%, and the average pixel accuracy was increased by 9.96%, the accuracy of the algorithm was increased by 11.52%, which verifies the effectiveness and feasibility of the proposed algorithm, and has certain application value in industrial applications.