混合像元分解_BNU_0425概要1.pptxVIP

陈学泓;relative terms;1. Introduction;What is spectral unmixing?;Why we need it?;Why we need it?;2. Linear spectral unmixing;Spectral mixture model;;;General unmixing procedure;1) Dimension Reduction;1) Dimension Reduction;2) Endmember Determination;;2) Endmember Determination;2) Endmember Determination;2) Endmember Determination;In interactive endmember determination, image endmember selection is achieved through an educated trial-and-error approach. An analyst has some knowledge of the field site or data set, and a set of objectives for conducting the analysis. Automated algorithms may employ statistics to capture variability, but their analytical determination may result in endmember estimates that satisfy some optimality criterion, but are physically unrealistic.;3) inversion;Residual error vs. retrieval error;Examples;;;Vegetation;Endmember;3. Further research;3.1 Nonlinear effect;;Nonlinear models;Issues in polynomial model ;Issues in polynomial model ;No strong collinearity effect in LSMA because only distinctive spectra with low correlation are selected as endmembers. However, since virtual endmembers are not selected using the true endmember extraction methods, collinearity of NSMA can be strong. The interaction term will be highly correlated with the other endmembers particularly when the spectrum of one endmember is relatively flat.;Issues in polynomial model ;In-situ measurement In order to construct a large number of testing cases for various degrees of collinearity, ten bands were randomly selected from the original data, producing 10 000 groups of ten-band spectral data. The rmses of constrained LSMA and pNSMA were evaluated. The LSMA performs better than NSMA in 61.5% of the cases. ;Linear vs. non-linear;3.2 Endmember variability;How to address endmember variability;How to address endmember variability;FISHER DISCRIMINANT based SMA;FISHER DISCRIMINANT based SMA;FISHER DISCRIMINANT based SMA;FISHER DISCRIMINANT based SMA;FISHER DISCRIMINANT based SMA;FI