The Study On Comparing Non-linear Unmixing Model And Linear Unmixing Model Of Mixed Pixels

The aim of the paper validates the validity of non-linear mixed unmixing model in classification of remote sensing data. This paper focuses on application of mixed pixels unmixing technology in classification of remote sensing data and compares the effects of nonlinear mixed model and linear mixed model in classification of remote sensing data and tries to solve the problem of mix pixels in remote sening data better and raise the precision of classification of remote sensing data.Because pixels in remote sensing image are typically mixed pixels due to both the limited resolution of sensors and the heterogeneous surfaces of ground covers. It's related to imaging mechanism, feature extraction and many other unavoidable problems. Meanwhile mixed pixels unmixing technology is one of the important approaches of classification of remote sensing data. There are two difficulties in mixed pixels unmixing technology. (1)Determine and pickup the terminal component endmembers (2)Solve the mixed pixels unmixing model.Firstly this paper does some pretreatment with ETM+ data so as to guarantee the farther analyzing and application. After determining the mixed pixels unmixing model, the precision is mostly affected by the choice of component endmembers. Secondly this paper do some research about the normal methods of choosing and determining the spectrum of component endmembers. Thirdly this paper detailedly describes the calculating course of the application of linear mixed unmixing and non-linear mixed unmixing model in classification of remote sensing data and compares the effects of non-linear unmixing model and linear unmixing model, The experiment proved that non-linear mixed unmixing model is more efficient than linear mixed unmixing model in classification of ETM+ data by raising 4% of precision and testify practicability and validity of non-linear mixed unmixing model in classification of ETM+ data.