Study on the automatic classification for land use/land cover in arid area based upon remotely sensed image cognition

Abstract

Traditional classification methods based on Bayes rule only use spectral information, whereas, other characteristics such as shape, size, situation and pattern are seldom taken into account to extract land use and land cover information. A new method based on spectral, contextual and ancillary information has been proposed in this paper to address to the problem of misclassification. The study area is located in an arid area of northern China. Based on eCognition software, A TM image and a DEM was utilized in this paper to investigate the effectiveness of the image-cognition based on classification method in land use/land cover classification of arid areas. The image was first segmented into a number of objects and then classified as 22 classes based on the spectral, shape, area, spatial position, pattern and context information with the fuzzy logic rules. Finally, the classification method has been proved to be effective and produced an overall accuracy up to 85.3% and a Kappa coefficient of 84%. The classification result suggests that this method is effective and feasible to classify the main types of ground objects in the large complex and arid area for land use survey.