Research On Image Enhancement And Automatic Defect Recognition Algorithm In Industrial Microfocus X-ray Imaging

In the case of mass production of devices,defect detection by human eyes alone will produce a high rate of missed detection and error detection,which can not meet the quality and speed requirements of factory production.Therefore,the automatic defect detection of workpiece by computer has become an significant research topic little by little.Because the attenuation degree of X-ray is different when X-ray penetrates the workpiece with different structures or materials with different thicknesses,the digital image obtained by the detector will contain the detailed information of the workpiece after X-ray penetrates the workpiece,and the technology will not cause damage to the workpiece and realize real nondestructive testing.However,the images collected by X-ray source contain a lot of background noise and poor detail contrast,which not only has poor rendering effect,but also affects the effect of automatic defect detection.Therefore,the enhancement and denoising of X-ray images are particularly important.In Thesis,micro focus X-ray source is used to image PCBA and lithium battery,C++ and OPENCV are used for image enhancement,and then OPENCV and HRNet are used to realize automatic defect detection of PCBA and lithium battery.The main research contents are as follows :1.An improved image enhancement algorithm based on gamma function is designed to achieve image enhancement and denoising.The new image enhancement algorithm contains two gamma functions,so as to realize the control of the stretching scale of the contrast between high and low gray areas of the image.By comparing and analyzing different image enhancement algorithms,the classical median filtering and multi-graph average method are finally selected to remove particle noise and random noise.2.Image-based morphological operation and breadth-first search algorithm(BFS)realized the detection of solder balls and bubbles in PCBA.Combined with the characteristics of image opening and closing operation,top cap and black cap,the multi-scale closing operation is used to extract the solder ball.Finally,the BFS algorithm is used to realize the segmentation of solder ball and the qualification rate discrimination.3.Defect detection of lithium-ion battery based on high-resolution network.After each convolution of HRNet,the high-resolution subnet of the image will exchange information with the low-resolution subnet to maintain the high-resolution characteristics of the image.Labelme was used to mark the key points at the end of the battery plate.The key points of the lithium-ion wound battery and the laminated battery were detected by the obtained model.Finally,the plate offsets of the wound battery and the laminated battery were obtained by the detection results and the conversion algorithm,so as to determine whether the wound battery and the laminated battery were qualified.