A Whitening Transformation Based Approach to One-class Classification of Remote Sensing Imagery

In this study,a whitening transformation based approach to one-class classification of remote sensing imagery is investigated.Only positive data are required to train the one-class classifier.Firstly,the image data is mapped to a new feature space using the whitening processing with all directions of the class of interest.Then a threshold is selected to make a binary prediction.A heuristic method of threshold selection is performed in the experiment of one-class classification.A series of values are set to the threshold based on standard deviation,and perform the one-class classification with each threshold value.The experiment shows that high accuracy is achieved with the threshold range from 3to 4standard deviations of the mean.Finally,the results of one-class classification with the threshold of 3standard deviations are compared to that of one-class support vector machine.The results indicate that the proposed method provides nearly the same accuracy of one-class classification as one-class support vector machine.The advantage of the proposed method is that it can use a constant threshold to extract various land types.