Semi-supervised Land Cover Classification with Category Label Extension based on Neighboring Pixels

抄録

It is necessary to recognize the categories of the earth surface correctly for the environmental problem and remote sensing area. For the above purpose, land cover is classified generally by using supervised methods. The disadvantage of supervised method is that sufficient amounts of training data are required. In this study, semi-supervised method with a limited number of training data is used to solve this problem.　Euclidean distance is used to reduce extension of training data as wrong category. With extended training data, maximum likelihood is used to classify by assigning categories for each area.