A self-adaptive study method for multi-parameters thresholds in AOI system

Hundreds of features extracted from object images need to be detected and every feature need to be evaluated by its threshold in AOI system. In order to cope with the study problem of multi-parameters thresholds during the solder joint surface qualify inspection in AOI system, a self-adaptive study method is proposed in this manuscript. Firstly, solder joint features are extracted from the solder joint image, which is divided into several key regions based on its features distribution. And the models of the solder joint types including good, excessive, poor, pseudo, shift, and tombstone are built based on extracting several key features of the solder joint. Secondly, considering two comparison ways of threshold parameters, a self-adaptive study method is proposed based on samples intelligent study by a designed closed loop feedback system. With the self-adaptive study method, all threshold parameters can approximate to their ideal values with increasing sample study. Thirdly, to evaluate the performances of the proposed method, several experiments are performed in the AOI system developed by authors. Experiment results show that threshold parameters can be obtained and converge to their ideal values quickly. Furthermore, with the help of the learned threshold parameters, the rate of correction inspection of AOI algorithm is up to 98.3%.