Research on Extraction and Assistant Classification of Remote Sensing for Texture Feature

Abstract

The classification precision of remote sensing image has always been one of the problems to each scholar. The traditional classification method is based on the spectral information. With the advancement of technology, the resolution of remote sensing image is gradually improving, and texture features included are getting rich, so adding texture characteristics to spectral characteristics for image classification can remedy the shortage of only relying on spectral characteristics. This paper uses experimental area of aerial image with 0.5m resolution in Datong, calculates fractal dimension using differential box-counting model, extracts the spatial texture features, and classifies precisely combining spectral characteristics in maximum likelihood method. Through comparing different classification results based on different characteristics, it show that classification accuracy based on combination of texture characteristics and spectral characteristics is more accurate (92% overall accuracy and kappa=0.91) than the one based on single spectral feature (88% overall accuracy and kappa=0.85) and texture feature (69% overall accuracy and kappa=0.65), which verifies the effectiveness of this method.