Abstract: One-class extraction from remote sensing imagery is a special method of classification,where users are only interested in recognizing one specific land type.The extraction of a specific class was studied based on nearest neighbor rule in this paper.Two aspects were considered,class partitioning and sample selection for each class.Firstly,the effect of data distribution partitioning is analyzed theoretically based on nearest neighbor in one-class classification.It is confirmed that the nearest neighbor classifier requires the data distribution to be partitioned into only two classes,namely the class of interest and the remainder.Secondly,as a two-class problem,the classification process was simplified,and the sample selection in nearest neighbor classification was performed in terms of both the spatial and the feature space.The experiments show that the specific class of interest can be well extracted from the remote sensing image with the proposed method.

Key words: One-class,Information extraction,Remote sensing,Nearest neighbor