Maximum Likelihood Method Modified in Estimating a Prior Probability and in Improving Misclassification Errors

Maximum Likelihood Method (MLM) has been one of the most traditional classification methods in remote sensing field, but its disadvantages have been also pointed out. While a prior occurrence probability gives a crucial effect to classification results, most of classifications have been conducted on an assumption that each a prior probability of land cover is equal because of insufficient a priori information. And as long as the class showing the highest likelihood is allocated to a pixel, misclassification errors are unavoidable. Authors modified method can estimate a prior probability through EM (Expectation Maximization) algorithm, applying a finite mixture model for a target image histogram. And misclassification errors can be overcome by data fusion model. Validation results demonstrate that data fusion model is effective to improve misclassification errors.