Discrimination problems for satellite images

摘要: Abstract This paper discusses the pitfalls of classifying satellite images using automatic classifiers. The discussion is illustrated by application quadratic and linear discrimination to high resolution data produced for National Remote Sensing Centre campaign. Within training set there good between most land uses, but lack agreement overall results two discriminators shows that measures accuracy based on within can be misleading. Cross-validation discriminator calculation site spectral variability also illustrate how usual calculations may give over-optimistic variability.