A DETAILED COMPARISON OF BACKPROPAGATION NEURAL-NETWORK AND MAXIMUM-LIKELIHOOD CLASSIFIERS FOR URBAN LAND-USE CLASSIFICATION

A detailed comparison of the backpropagation neural network and maximu m-likelihood classifiers for urban land use classification is presente d in this paper, Landsat Thematic Mapper images of,Tucson, Arizona, an d Oakland, California, were used for this comparison, For the Tucson i mage, the percentage of matching pixels in the two classification maps was only 64.5%, while for the Oakland image it was 83.3%. Although th e test site accuracies of the two Tucson maps were similar, the map pr oduced by the neural network was visually more accurate; this differen ce is explained by examining class regions and density plots in the de cision space and the continuous likelihood values produced by both cla ssifiers, For the Oakland scene, the two maps were visually and numeri cally similar, although the neural network was superior in suppression of mixed pixel classification errors, From this analysis, we conclude that the neural network is more robust to training site heterogeneity and the use of class labels for land use that are mixtures of land co ver spectral signatures, The differences between the two algorithms ma y be viewed, in part, as the differences between nonparametric (neural network) and parametric (maximum-likelihood) classifiers. Computation ally, the backpropagation neural network is at a serious disadvantage to maximum-likelihood, taking nearly an order of magnitude more comput ing time when implemented on a serial workstation.