#### LARS Tech Report Number

030772

#### Abstract

The utilization of minimum distance classification methods in remote sensing problems, such as crop species identification, is considered. Minimum distance classifiers belong to a family of classifiers referred to as sample classifiers. In such classifiers the items that are classified are groups of measurement vectors (e.g. all measurement vectors from an agricultural field), rather than individual vectors as in more conventional vector classifiers.

Specifically in minimum distance classification a sample (i.e. group of vectors) is classified into the class whose known or estimated distribution most closely resembles the estimated distribution of the sample to be classified. The measure of resemblance is a distance measure in the space of distribution functions.

The literature concerning both minimum distance classification problems and distance measures is reviewed. Minimum distance classification problems are then categorized on the basis of the assumption made regarding the underlying class distribution.

Experimental results are presented for several examples. The objective of these examples is to: (a) compare the sample classification accuracy (% samples correct) of a minimum distance classifier, with the vector classification accuracy (% vector correct) of a maximum likelihood classifier; (b) compare the sample classification accuracy of a parametric with a nonparametric minimum distance classifier. For (a), the minimum distance classifier performance is typically 5% to 10% better than the performance of the maximum likelihood classifier. For (b), the performance of the nonparametric classifier is only slightly better than the parametric version. The improvement is so slight that the additional complexity and slower speed make the nonparametric classifier unattractive in comparison with the parametric version. In fact disparities between training and test results suggest that training methods are of much greater importance than whether the implementation is parametric or nonparametric.

#### Date of this Version

1-1-1972