Application research on the method for extracting soybean covered areas based on the pixel unmixing

When estimating the yield of large area crops from the low and middle resolution images of the remote sensing satellites, the influence of the mixed pixel should be considered in order to improve the accuracy of the soybean covered areas extraction. This paper selected Lishu County of Jilin Province as a test area, applying the Linear Spectral Model (LSMM) to classify TM images and evaluate soybean covered areas. The result was compared with the interpretation result from Quickbird image. According to the quantity accuracy assessment method, the classification accuracy of soybean covered areas reached 92%. This paper also extracted soybean covered areas by the maximum likelihood supervised classification and the isodata unsupervised classification. The corresponding accuracies are 87% and 84%, respectively. Result shows that the pixel unmixing techniques can improve the classification accuracy of extracted soybean covered areas comparing with other quantification methods of remote sensing.