Design and Verification of Visual Detection Algorithm for Mini LED Chip and Panel Defects

To enhance the quality and speed of defect detection in Mini LED chip panels, we have focused our research on defect detection methods using vision algorithms and strategies for algorithm acceleration. This paper introduces detection methods based on threshold segmentation and dynamic threshold segmentation, and constructs an improved algorithm architecture by combining the advantages of both. Threshold segmentation is suitable for detecting targets with significant contrast differences from the background, whereas dynamic threshold segmentation can detect defects with less obvious contrast. Additionally, for the high-speed design of the algorithm, the use of parallel computing through threading can effectively reduce computation time. Since Mini LED chip panels typically require high-resolution vision systems, which increase the time required for image acquisition and processing, a slicing algorithm is designed to reduce image size and thus improve speed. Furthermore, CUDA acceleration strategies can further enhance the processing speed. Finally, chip alignment experiments are conducted on a multi-degree-of-freedom spatial pose adjustment system to verify the feasibility of the vision algorithm as a feedback adjustment method.