Mixed Pixel Unmixing And Graphic Recovery Based Thin Clouds Elimination Research In Remote Sensing

Clouds are common in remote sensing images obtained by satellite. Thisphenomenon has many disadvantages such as low efficiency in the usage of remotesensing data. As to thick clouds, the ground information has been completely blockedand can‘t be recovered; as to thin clouds, the ground information has only beenjammed rather than been completely Locke, so it‘s possible to remove thethin-clouds-interference and recover the remote sensing data. From the principle ofremote sensing data acquisition, single pixel is the spectral composite of the groundspectral information. In many cases, the pixel reflects more than one kinds of surfacefeatures——usually, this kind of pixel which composites a variety kinds of surfacefeatures‘spectral information is called mixed pixel, the surface features which isreflected by the mixed pixel are called endmembers and each endmember‘scontributory percentage in a mixed pixel is called abundance.In the area covered with thin clouds, remote sensing data pixel is considered tobe the reflection of the ground spectral information and the thin clouds‘spectralinformation. So these pixels can be seen as the mixture of thin clouds and severalsurface features. Therefore, we can use the mixed pixel unmixing method in the fieldof image processing to calculate a variety of endmembers and theirs abundance, suchas the ground surface features and thin clouds. Subsequently, through the abundanceadjustment–removing the thin clouds‘endmembers and adjusting the otherendmembers‘abundance (the other surface features‘abundance sum to one), the useof surface features‘spectral curve to synthesize free-cloud mixed pixel, thuscompleting the remote sensing data‘s recover in thin cloud area.This paper uses an efficient method in the mixed pixel unmixing field-theminimum volume constrained nonnegative matrix factorization to calculate the abundance of their corresponding mixed pixel endmember, and then use theabundance adjustment method to recover the remote sensing data. This paper uses theIDL platform to programming and testing the method above through TM, ASTER andHyperion data. This paper has achieved some success and also gained manyexperiences. May these achievement can provide some reference to others.