Neural networks versus nonparametric neighbor-based classifiers for semisupervised classification of Landsat Thematic Mapper imagery

Semisupervised classification is one approach to converting multiband optic al and infrared imagery into landcover maps. First, a sample of image pixel s is extracted and clustered into several classes. The analyst next combine s the clusters by hand to create a smatter set of groups that correspond to a useful landcover classification. The remaining image pixels are then ass igned to one of the aggregated cluster groups by use of a per-pixel classif ier. Since the cluster aggregation process frequently creates groups with m ultivariate shapes ill suited for parametric classifiers, there has been re newed interest in nonparametric methods for the task. This research reports the results of an experiment conducted on six Landsat Thematic Mapper imag es to compare the accuracy of pixel assignment performed by four nearest ne ighbor classifiers and two neural network paradigms in a semisupervised con text. In all the experiments, both the neighbor-based classifiers and the n eural networks assigned pixels with higher accuracy than the maximum-likeli hood approach. There was little substantive difference in accuracy among th e neighborhood-based classifiers, but the feedforward network was significa ntly superior to the probabilistic neural network. The feedforward network classifier generally produced the highest accuracy on all six of the images , but it was not significantly better than the accuracy produced by the bes t neighbor-based classifier. (C) 2000 Society of Photo-Optical Instrumentat ion Engineers. [S0091-3286(00)03807-1].