Abundance Extraction of End-Members of Forest Based on Linear Mixed Model - A Case Study of Meijiang Basin in China

Because of ignoring mixed pixels, classification errors will be consequentially generated based on generally supervised classification by per pixel. This paper takes Meijiang River Basin as the investigative object. After MNF and PPI, main forest end-members (broadleaf forest end-member, coniferous forest end-member and low herbage end-member) abundances maps were obtained with Linear Spectral Model, combining with reality and a special device which helped us interactive selection. Results showed that the soft classification is an effective method of improving the precision of remote sensing classification to a certain extent.