Land Use Classification of TM Imagery in Hilly Areas: Integrationof Image Processing andExpert Knowledge

摘要

Improvement of the classification accuracy is one ofrnthe major concerns in the field of remote sensing applicationrnresearch in recent years. Previous research shows that the accuracyrnof the conventional classification methods based only onrnthe original spectral information were usually unsatisfied andrnneed to be refined by manual edit. This present paper describesrna method of combining the image processing, ancillary datarn(such as digital elevation model) and expert knowledge (especiallyrnthe knowledge of local professionals) to improve thernefficiency and accuracy of the satellite image classification inrnhilly land. Firstly, the Landsat TM data were geo-referenced.rnSecondly, the individual bands of the image were intensitynormalizedrnand the normalized difference vegetation indexrn(NDVI) image was also generated. Thirdly, a set of samplernpixels (collected from field survey) were utilized to discoverrntheir corresponding DN (digital number) ranges in the NDVIrnimage, and to explore the relationships between land use typernand its corresponding spectral features. Then, using the knowledgerndiscovered from previous steps as well as knowledgernfrom local professionals, with the support of GIS technologyrnand the ancillary data, a set of conditional statements werernapplied to perform the TM imagery classification. The resultsrnshowed that the integration of image processing and spatialrnanalysis functions in GIS improved the overall classificationrnresult if compared with the conventional methods.