篇名 | Detection and Segmentation of Defects in Industrial CT Images Based on Mask R-CNN |
---|---|
卷期 | 31:6 |
作者 | Jun-Nian Gou 、 Xiao-Yuan Wu 、 Li Liu |
頁次 | 141-154 |
關鍵字 | defect detection 、 defect segmentation 、 industrial CT image 、 mask R-CNN 、 EI 、 MEDLINE 、 Scopus |
出刊日期 | 202012 |
DOI | 10.3966/199115992020123106012 |
For defect detection and segmentation of industrial computed tomography (CT) images, existing methods regard them as two separate tasks, that is, the detection and the segmentation are not integrated. This paper improves mask regions with convolutional neural networks (Mask R-CNN), and the improved network is used to complete the detection and segmentation of industrial CT image defects at the same time. In this study, the three most common defects in industrial CT images are cracks, bubbles, and slags, which are taken as the research objects. The experimental results showed that the network can achieve a detection mAP value of 0.98 and a segmentation accuracy of 0.96 for three typical CT image defects (12 × 12 pixels ~ 35 × 35 pixels). The method proposed in this paper has high accuracy, in both the defect detection and segmentation of industrial CT images. At the same time, the algorithm also has good robustness and generalization ability. It can be applied to the detection and segmentation of various defects and has great practical significance.