MRF-SegFormer : A Magnetic Leakage Defect Segmentation Network Based on Segformer

Traditional segmentation networks have low Accuracy in segmenting defects in magnetic flux leakage images, and are prone to small defects being missed. This has a significant impact on evaluating the degree of defect harm using the results of magnetic flux leakage image defect segmentation. This article proposes a high-precision decoder based on cross feature fusion, which is more suitable for magnetic flux leakage defect detection, on the basis of the SegFormer network, to replace the traditional multi-layer perception (MLP) decoder. The experiment TA results show that the model increases the MIoU value of the traditional MLP decoder from 88.21% to 89.63% with a small number of additional parameters