Adaptive sorting by prototype populations

Abstract

An algorithm for unsupervised adaptive sorting is presented, based on a finite number of ‘prototype populations’, with distinctly different feature distributions, each representing a typically different source population of the inspected products. Updated feature distributions, of samples collected from the currently sorted products, are compared to the distributions of the stored prototype populations, and accordingly the system switches to the most appropriate classifier. Although the goal is similar to the objectives of previously proposed ‘Decision Directed’ adaptive classification algorithms, the present algorithm is particularly suitable for automatic inspection and classification on a production line, when the inspected items may come from different sources. The practical feasibility of the approach is demonstrated by two synthetic examples, using Bayes classifiers. This is followed by an applied example, wherein two prototype populations of apples are sorted by size, derived by machine vision. It is shown that misclassification by adaptive classification is reduced, in comparison to non-adaptive classification.