PCB Solder Joint Defects Detection And Classification Based On Computer Vision

PCB(Printed Circuit Board) has the characteristics of high density, high precision, and high reliability. With the number of PCB layers increasing, the density of PCB becomes higher, and the size of PCB is also smaller, which makes the inspection of PCB solder joint defects more difficult.In order to meet the Gree’s demand for the inspection of PCB solder joint defects, we analyze the related defects, and propose a practical solution based on computer vision for the detection and recognition of four typical defects including short circuit, solder skips, poor and excessive solder. The main contributive work of this thesis can be summarized as follows:(1) Color channel selection for the PCB image.The PCB solder joint area hold different image characteristics in different color space and different color channel. In order to make the defects detection and recognition more efficient, this thesis first carried out the selection of color space and color channel for the PCB image. The experimental results show that the R channel of the RGB color model is more suitable for the description of PCB image.(2) Detection and recognition of PCB solder joint defects based on mathematical morphology.In this method, we first filter the PCB image extracted from R channel. We then convert the filtered image into binary image, and do feature extraction based on connected component labeling and binary morphological filter. Finally we construct the rule for the detection of the PCB solder joint area and the defects type recognition. As the parameter setting in the aforementioned rule is related to the absolute area value of PCB solder joint area, the area threshold should be reset when the relative position between the surveillance camera and the inspection line is changed.(3) Detection and recognition of PCB solder joint defects based on image registration.In this method the PCB image without welding is first captured as the standard reference image. Then the position corresponding to the welding point and the region of interest area can be set in this reference image. By means of automatic image registration, the PCB solder joint area can be extracted from the corresponding input image and geometrically rectified based on the reference ROI region. Finally we construct the rule for the detection of the PCB solder joint area and the defects type recognition. Different from the first method, we introduce the relative area threshold. Instead of resetting a new threshold, we just recapture the standard reference image when changing the relative position between the surveillance camera and the inspection line.The experimental result shows that the method based on image registration can meet the current demand of the Gree’ production.