Accurate defect detection via sparsity reconstruction for weld radiographs

摘要

Detecting defects in weld radiographs is an important research topic in the field of industrial non-destructive testing. Many computer-aided detection techniques have been designed for detecting defects. However, these techniques are mainly used to detect specific defective types. They cannot be applied to detect diverse types of defects, which is a difficult task because the number and types of defects in weld radiographs are generally unknown in advance, and different defects may exhibit different visual properties in shapes, sizes, textures, contrasts and positions. Inspired by the experienced workers' visual inspection mechanism, this paper develops a novel framework to detect diverse types of defects from X-ray images. In the framework, a large number of normal X-ray images are firstly collected to serve as "workers' experience" and guide the defect detection. Then, a dictionary is learned from the collected normal set. It can selectively reconstruct the background and the weld region of a test image while suppressing defective regions via sparsity reconstruction. By computing the difference image between the test image and its reconstructed image, flaws are well highlighted as the reconstruction residuals and separated from the difference image. Extensive experiments have shown that the proposed technique detects diverse defects more accurately compared with the state-of-the-art methods.