A Hierarchical Object Oriented Method for Land Cover Classification of SPOT 5 Imagery

Land cover classification with a high accuracy is necessary, especially in waste dump area, accurate land cover information is very important to eco-environment research, vegetation condition study and soil recovery destination. Funded by the international cooperation project Novel Indicator Technologies for Minesite Rehabilitation and sustainable development, a hierarchical object oriented land cover classification is produced in this study. The ample spectral information, textural information, structure and shape information of high resolution SPOT 5 imagery are used synthetically in this method. There are two steps in object oriented information extraction: image segmentation and classification. First, the image is segmented using chessboard segmentation and multi-resolution segmentation method. Second, NDVI is used to distinguish vegetation and non-vegetation; vegetation is classified as high density vegetation, middling density vegetation and low density vegetation using spectral information, object oriented image texture analysis; non-vegetation is classified as vacant land and main road using length/width. Accuracy assessment indicate that this hierarchical method can be used to do land cover classification in waste dump area, the total accuracy increases to 86.53%, and Kappa coefficient increases to 0.7907. Key-Words: hierarchical land cover classification, NDVI, object oriented texture analysis, waste dump opencast coalmine area, SPOT 5