Land Use/Cover Information Extraction Using Remote Sensing and GIS

摘要

Landsat TM data of Fengning County, Hebei Province, including seven bands color composition, image geometric correction, mosaic and subset, was first processed in 1999. Principal component analysis of Landsat TM image was applied. We selected inverse PCA image as information source for supervised classification. In ERDSA IMAGINE 8.5 software, signatures of typical classes were gathered using AOI polygon and seed growth properties. To improve classification accuracy, we selected remote sensing image in July 1987, thematic data in GIS, such as relief, geology, soil, soil erosion and land use map, DEM and field survey GPS data of typical classes as reference information when selecting training samples. Spectral response curve, error matrix and feature space were selected to evaluate the quality of training samples. It showed that the result of selected training samples evaluated by the above three measures was satisfied. We reselected and modified training samples many times after evaluating them, until accuracy of all samples reached above 85 percent. We used maximum likelihood method to make supervised classification. The final classification accuracy was 84.62 percent.