Color Grading of Cotton Part II: Color Grading with an Expert System and Neural Networks

Abstract

In this part of the series, two color grading systems are developed using an expert system and neural networks. Both grading systems have two modes of operation— classification and training. In the training mode, the expert system can be trained by a statistical method based on Bayes' theorem or a genetic algorithm. For the neural network approach, the grading system can be trained by a back-propagation algorithm or a probabilistic neural network. Using 100 cotton samples from the USDA, the agreement between classer and HVI grading can be improved from the original 50% to 86-100% depending on the training method and the training samples. The relative contributions of each measurement on color grading are also investigated using stepwise discriminate analysis.