A study of two methods for accuracy assessment to RS classification

RS accuracy assessment for RS classification is a method by which reliability and variability of the result of RS classification are quantificationally described. The reliability analysis is a process of calculating classification accuracy at a probability level while the variability analysis is to estimate the dependability range of the classification accuracy. With a point sample the AEFD (Accuracy Estimation for Feature Discrimination by RS) and the Error Matrix were selected to assess the RS classification results derived from several RS images spatially distributed on an experiment region in Xin Jiang province. The point samples which consist of many point features of GIS contain two kinds of factors which are called referenced data and assessed data. Based common geographic coordinate the point features connect the two factors. With GPS in fields sampling point features acted as a main method was used to obtain the point sample in this experiment. The experiment result presents that the AEFD is easy to calculate and the dependability range of the classification accuracy can be estimated when the sample size is more than 50. The Error Matrix has several statistical indexes which illustrate the situation of RS classification from several aspects without probability; meanwhile, the process of calculation is comparatively difficult.