Magnetic column defect recognition based on Deeplab V3+

In order to solve the defect detection problem caused by the loss of detail information and low segmentation accuracy of Deeplab V3+ model, an improved Deeplab V3 + semantic segmentation model was proposed to detect magnetic column defects. Firstly, in order to reduce the model's parameter quantity, Mobilenet V2 is updated and used as the skeleton network of Deeplab V3+. Secondly, the attention module is added in the block of Mobilenet V2 network to strengthen the processing of details. Finally, the ASPP module is replaced with the D-ASPP module, which obtains richer semantic information and covers a bigger scale more densely. The experimental findings demonstrate that the novel method's mIoU value is 79.25%, which is 2.88% higher than that of the original Deeplab V3 + algorithm, and the defect recognition precision reaches 96.87%. Parameter quantity is decreased by 52.95%, operation time reduced by 27.8%.