Pattern classification system for the automatic analysis of paper for recycling

Abstract

The objective of this paper is the development of a pattern classification system for the automatic analysis of the composition of samples from paper for recycling. The system uses a colour video camera and dynamic scales as sensors. An overall number of 26 features of the categories weight, shape, colour, texture and amount of optical brighteners are extracted to distinguish between ten classes of paper for recycling. Five classifiers of different types are trained and tested with industrial samples. During testing a piece of paper is assigned to the class most common among the five classifiers, but the classification success rates lie under 50% for some classes and are therefore not acceptable. For this reason, the system is modified to discriminate only between six of the ten classes. For those classes, the classification success rates lie between 94% and 100% which is suitable for the analysis of samples from paper for recycling.