Study Of Pixel Unmixing Technology And Method Based On MODIS Remote Sensing Data

The lower the remote sensing images' spatial resolution, the greater the probability for one pixel containing a wide range of different goals. Currently, the popular medium-high resolution remote sensing data, which is characterized by higher data prices, access difficulties, though it has high measurement accuracy; and the low-resolution data, such as MODIS remote sensing data, is extensively covered and low-cost, However, because of its low spatial resolution, remote sensing imagery existence of a large amount of mixed pixel, traditional remote sensing image classification is not considering in this issue. As a result, smaller types of features have been assigned to the wrong category. However, the pixel unmixing study used data is NOAA/AVHRR or Landsat TM/ETM remote sensing data, but the mixed spectral analysis based on MODIS remote sensing data is little. To solve the above problem, this paper chooses Zhengzhou as an example, selects the linear mixture model to analyses mixed pixel in MODIS remote sensing data.First, this paper discusses the MODIS data formats and pre-processing, analyses and summarizes the commonly used pixel unmixing models. Through MNF transform, scatter plots and the introduction of pure pixel index to define the end members, and uses the least pixel contain method and geographic space images selection method to gets the end members' reflectance. Then the two methods' results of the end members' reflectance are substituted to linear mixture model equation, constrained and not constrained least-squares method will be used individually, there will be each end member type of percentage (abundance) and the RMS error in the result plots. This shows that the results of pixel unmixing which use several methods all have the RMS mean less than 0.003. Compared with an interpreted TM remote sensing data, result to a comprehensive evaluation of geospatial binding law the outcome is best. The outcome resulted from a comprehensive evaluation of geospatial binding law is best. Finally, the traditional supervised classification and unsupervised classification of the original image classification, respectively classify the original image with the traditional supervised classification and unsupervised classification. The results show that the unsupervised classification has the worst effects, and the supervised classification is better than non-supervised classification. But the two classifications results failed to meet the requirements. To enable more precise classification, it is needed to use mixed pixel decomposition method which has better comparative results. The results that the use of mixed-pixel decomposition technology MODIS land remote sensing data classification can achieve better results show MODIS data can be effectively applied to remote sensing dynamic monitoring, land cover classification.