| There are too much mixed pixels in Remote Sensing images, now. It is urgent for Xinjiang province and national cotton monitoring system to recognize cotton from mixed pixels and get cotton information from them. According to this, the author having research on mixed pixel unmixing, and choosing LSMM(linear spectral mixture model) to unmix mixed pixel. About the endmembers, they are picked up from Remote Sensing images, they are also got from field and laboratory surveying. The endmembers picked up from Remote Sensing images are called image endmember, endmembers got from indoor or field surveying named reference endmember. Reference endmember is standard spectrum, atmospheric or sunny condition has little effect on it, and it is chosen to unmix mixed pixels in this paper. It is DN(digital number) at the height of satellite level in the Remote Sensing image, only after it is converted to ground level reflectance, the Remote Sensing image and reference endmember can be used to unmix mixed pixels. 6s(second simulation of the satellite signal in the solar spectrum) software is used in the paper to do atmospheric calibration to the Remote Sensing image, the software changes the DN to reflectance. Graphic models are written by module modeler in ERDAS IMAGINE to carry out unmix mixed pixel, at the end, the task of unmix mixed pixels in the experiment fields is completed successfully.In the last years, the author have checked a great deal of domestic and international literature data, at the foundation of earnest study, analysis and summarization the experience of many scientific researcher, with the cordial help of the tutor and other teachers, with precise research attitude, diligent work spirit, the author have completed the reserved research contents. Now summarized as follows:(1) Linear spectral mixed model, fuzzy supervised classification model and artificial neural network model etc are used in mixed pixel unmixing, the models have its own advantages and fields of application respectively, among which, linear spectrum mixed model has simple structure, distinct physical meaning, so it is used often. Its application fields, including information gathering, land cover classification, water pollution assessment by Remote Sensing and Surface component temperature inversion.(2) We measure endmember spectrum by spectrometer, and TM image is atmospheric calibrated. measured Spectrum is continuous, but TM image has discrete band ; measured spectrum file is entered into ENVI as spectral library file, then the spectral library file is resampled to TM bandwidth, so measured spectrum has the same bandwidth as tm image. TM image is atmospheric calibrated by 6S software, In the software, many unknown parameters are assessed reasonably. Then TM image and measured endmember spectrum are ready for mixed pixel unmixing.(3) Linear spectral mixed model is used to unmix mixed pixels, we measured crops acreage to test the accuracy of mixed pixel unmixing. Modeler Maker is a object-oriented model of language environment in ERDAS, in the environment, we draw flow chart on a page using intuitive graphical language, and define graphics as input data, output data and operating function individually, then a spatial graphic model arise. we draw a graphic program in the modeler maker based on principle of linear spectrum mixed model, run the program, we get results of mixed pixel unmixing. We select some pixels randomly to test the accuracy, the maximum error is 0.09. we count cotton acreage in different fields, tested by measured cotton acreage, accuracy is above 90%.
|
PDF Full Text Request