Research on visualization of metal magnetic memory testing based on computer vision and deep learning

Metal magnetic memory (MMM) testing is widely used in the non-destructive testing (NDT) field of steel structures due to its rapid detection speed and simple operation. During the detection process, the Hall probe, which is used to detect the magnetic field, moves along the workpiece surface. Each scan can only obtain the magnetic field distribution along a single scanning path. In practical engineering applications, due to the large area of the detection object, high requirements for scanning efficiency, and the constraints of the on-site environment, the spacing between scanning lines is typically large. In such cases, the crucial structural information (such as the magnetic field distribution information at the crack edge) is frequently overlooked or even completely lost during the sparse scanning process, resulting in a decrease in the recognition rate for two-dimensional defects. Based on the principle of MMM detection, this paper conducts research on the reconstruction of the leakage magnetic field component surface under the MMM effect. By calculating the magnetic field intensity values obtained at each scanning point and mapping them to a two-dimensional plane based on the sampling positions, a two-dimensional image is formed. On this basis, an edge-aware super-resolution network is proposed to perform super-resolution (SR) on the magnetic field distribution image, and the U-Net is used to achieve the segmentation and recognition of crack defects.