Visual inspection system for crack defects in metal pipes

Surface cracks pose a hidden danger to metal pipes application. Several technical issues exist in crack detection in metal pipes, such as overexposure, low accuracy, and low detection speeds. To address the aforementioned key technical issues, this study proposes an image acquisition system with multiple lighting fusion and a metal-pipe crack defect segmentation model. These methods achieve the objective of acquisition and recognition of surface cracks in metal pipes by combining annular and coaxial light and introducing dual attention and boundary refinement modules. We contribute a surface defects on metal pipes dataset. Empirical evidence demonstrates that, within our proposed dataset, the methodology presented in this paper surpasses 20 state-of-the-art methods in terms of detection accuracy, while achieving a detection speed of 161.9 fps. Furthermore, in comparisons on public datasets, this study’s method outperformed 12 state-of-the-art approaches in achieving superior detection accuracy. The code for the metal-pipe crack defect segmentation model is located in https://github.com/zz-ux/Semantic-Segmentation-Network-for-Surface-Cracks-in-Metal-Pipes.