Statistical classification methods, such as the Bayesian classifier, can pr
ovide optimal classification but their performance depends heavily on the a
ssumption of normality of the input data. Artificial intelligence (Al) appr
oaches, on the other hand, entail less stringent assumptions about the stat
istical characteristics of the input data. Hence, the neural network and fu
zzy logic classifiers are expected to perform better than the Bayesian clas
sifier for a given data set. This paper describes steps involved in the dev
elopment of an optimal neural network classifier and a fuzzy classifier for
sorting apples using the selected image features as the input variables. P
erformance of the Al classifiers developed was compared with that of the Ba
yesian classifier using the same data set. The fuzzy classifier (80%) perfo
rmed as well as the Bayesian classifier with linear discriminant functions
(79%), whereas, the neural classifier performed better (88%). (C) 2001 Sils
oe Research Institute.