A Variable Attention Nested UNet++ Network-Based NDT X-ray Image Defect Segmentation Method
Jiayin Liu, Jae Ho Kim
Abstract
In this paper, we describe a new method for non-destructive testing (NDT) X-ray image defect segmentation by introducing a variable attention nested UNet++ network. To further enhance the performance of the faint defect extraction and its clear visibility, a pre-processing method based on pyramid model is also added to the proposed method to effectively perform high dynamic range compression and defect enhancement on the 16-bit raw image. To illustrate its effectiveness and efficiency, we applied the proposed algorithm to the X-ray image defect segmentation problem and carried out extensive experiments. The results support that the proposed method outperforms the existing representative techniques in extracting defect for real X-ray images collected directly from industrial lines, which achieves the better performance with 89.24% IoU, and 94.31% Dice.