Research On Technology About Hyperspectral Image Classification Of Mixed Pixels

Hyperspectral remote sensing plays an important role in remote-sensing area's development in Domestic and foreign. Hyperspectral remote sensing is the high spectral resolution remote sensing science and technology, its reliance on general science based testing (spectroscopy), because of its high spectral resolution,it can provide more information of earth's surface, And received more attention in the field of remote sensing applications. Hyperspectral remote sensing bring a great change to remote sensing field, it makes the reality that the objects which can not be detected in the wide-band spectrum can be detect.Hyperspectral image separation and object identification is one of the main applications of remote sensing technology. Remote sensing image recording information unit is called pixel, it records the corresponding spectral characteristics of surface material and information. Remote sensing image pixel corresponding to the surface material is rarely covered by a single type of surface composition, generally a mixture of several surface features. Thus the image pixel spectra characteristics of the spectrum is not only one, but several surface features reflect the mixture of spectral features, it has brought great impact on the accuracy of remote sensing applications. Mixed pixels either directly attributable to a typical surface features is wrong, since it at least not entirely belong to such a typical object. If each mixed pixel can be broken down into its component cover types (often referred to as end-member components), and each component calculated percentage of total pixel content classification more precise, because of mixed pixels arising from the ownership of the wrong and incorrect sub-problems will be solved, we call this process decomposition of mixed pixels.Mixed pixel problem is not only a obstacle to quantitative remote sensing technology's further development, but also seriously affect the effect of computer processing or computer technology in the field of remote sensing application. Most image classification algorithm does not consider this phenomenon, but as per spectrum using the statistical characteristics between the pixel classification. Mixed pixel decomposition technique consider this phenomenon, which is not only given composed of pixels around the table cover abundance classes, but also give the classification image. Today, commonly used spectral mixture analysis algorithm are as the following: least square, single physical volume,and so on. This paper will analysis of several algorithm deeply and make trues, later, use models complete the experimental data's testing and analysis.