Hierarchical Classification for Imbalanced Multiple Classes in Machine Vision Inspection

Abstract

Quality inspection based on machine vision is a multi-classification for imbalanced samples from many minority default-classes and a majority normal-class. Traditional methods seeking classification accuracy over a full range of instances are not suitable to deal with this case, since they tend to classify all samples into the majority class, usually the less important class. This paper proposed a hierarchical classification method that a simple bi-classifier with less features input made out most normal-class samples with a permitted low error rate for minority samples, then the rest less imbalanced samples were learned to establish a complicated multi-classifier with more features input. In classification after learning, two classifiers worked parallel and the simple classifier of the first layer can end the second one when normal-class result has been got. Comparative experimental results showed that this approach could effectively improve learning performance and accelerate classification speed.