Bootstrap sampling technique applied to the PCB golden fingers defect classification study

Abstract

Surface defects classification techniques commonly used in the literature are statistical methods (such as the Bayesian classifier, the linear discrimination function classifier, the minimum distance classifier and the nearest neighbour classifier) and non-statistical methods (such as neural networks and fuzzy logic). The PCB industry in Taiwan is already a small volume with a large variety type of production. If the number of defect samples is too few, it is difficult to apply any of the above-mentioned techniques for classification. The Bootstrap sampling technique is proposed for the generation of enough samples to determine population parameters for golden finger defect classification. An experiment was conducted to demonstrate the application of the technique. The tree classifier was used as the classifier in this study. The results showed that the classification correctness reached 97.87%, as opposed to 84.46% when other classifiers were used. The results provide an effective solution for the defects classification problem with a small sample situation.