“…1. Specifically, the procedure followed is the next: 1) The CMOS camera consists of four parameters, that is, the pixel size matrix is 3072*2048, framerate is 17 fps , exposure time is in the [27 2.5 ] ss , , and the pixel size is 2.4 m ; 2) The lens model is MVL-HF1628M-6MP; 3) Considering the illumination uniformity, LED (Light-Emitting Diode) is best for our study and the placement angle is 45° (Note that the dark field illumination technique can also be applied effectively, such as the 3D laminar flow-based method in [33]); 4) The image acquisition card model is MV-8808 because of its excellent compatibility; 5) The external trigger system includes I / O (Input/Output), motion control, level conversion unit, etc. A Flow-chart of the proposed system is shown in Fig.2, this framework is described as follows: 1) The framework consists of five parts, that is, input defect images, data preprocessing, edge detection, defects location, and output results; 2) The raw images with different defects are as the input; 3) For data preprocessing, four types of noises can be eliminated effectively, including common noises and special noises; 4) To detect the edge of metal components, both Laplacian-based edge detector and Gaussian-based Laplacian method are proven to perform well; 5) Surface defects location is the key technology in this proposed system, and its basic idea is to mine incremental properties and parallel properties; 6) Finally, detection results can be outputted, including the types of defects, recognition accuracy, the size of defects, and the coordinates of defects.…”