SVM Based MODIS Land Cover Classification Feature Select And Apply

MODIS (Moderate Resolution Imaging Spectra radiometer) is an important data getting instrument of information onboard the Terra and Aqua satellites. These data as the substitute of NOAA-AVHRR data are familiar with people. The data have been mainly used in dynamic monitoring of natural disaster, changes in landscape and vegetation, global net primary productivity, ecological environment, climate changes, and ocean.Support Vector Machines (SVM) are a novel machine learning method proposed in 1990s, they have excellent performance in resolving small-sample, non-linear, high dimension pattern recognition problem, which becoming the hotspot of machine learning field. In this paper, based on SVM, the extraction of classification features in Land Cover is studied. The results demonstrate that using multi-dimension info is super than usual three-dimension info(R, G, B), especially the multi-time muti-dimension info; the results also demonstrate that derived info such as vegetation index, water index can't improve the classification precision greatly, but they can represent the time information.