Multi-scale uncertainty evaluation of remote sensing image classification

Abstract

Remote Sensing (RS) image classification is one of the most important ways to extract thematic information, which used broadly in many fields. More and more attention has been drawn on the data quality recently. It is crucial to assess uncertainty of RS image classification, but the methods used so far for this task cannot provide information fully and completely. Based on information theory and rough set theory, the paper proposed a multi-scale evaluation (MSE) method, which is based on pixel scale, feature type scale and whole image scale, to realize the uncertainty evaluation of classification. The result of TM RS image classification was experimented on its accuracy of evaluation. At the same time, the static visualization of multi-scale evaluation for three classified images was carried out. Test result shows that the uncertainty evaluation by the multi-scale method is convenient for users to understand the uncertainty of classified image on pixel scale, different feature type scale and whole image scale, and it is also useful for the application of classified image.