HIERARCHICAL MAXIMUM-LIKELIHOOD CLASSIFICATION FOR IMPROVED ACCURACIES

Among the supervised parametric classification methods, the maximum-li kelihood (MLH) classifier has become popular and widespread in remote sensing, Reliable prior probabilities are not always freely available, and it is a common practice to perform the MLH classification with eq ual prior probabilities, When equal prior probabilities are used, the advantages in MLH classification mag not be attained, This study has e xplored a hierarchical pixel classification (HPC) method to estimate p rior probabilities for the spectral classes from the Landsat thematic mapper (TM) data and spectral signatures, The TM pixels were visualize d in multidimensional feature space relative to the spectral class pro bability surfaces, The pixels that fell within more than one probabili ty region or outside all probability regions were categorized as the p ixels likely to misclassify, Prior probabilities were estimated from t he pixels that fell within spectral class probability regions, The pix els most likely to be correctly classified do not need extra informati on and were classified according to the probability region in which th ey fell, The pixels likely to be misclassified need additional informa tion and were classified by MLH classification with the estimated prio r probabilities, The classified image resulting from the HPC showed in creased accuracy over three classification methods. Visualization of p ixels in multidimensional feature space, relative to the spectral clas s probability regions, overcome the practical difficulty in estimating prior probabilities while utilizing the available information.