Land Cover Classification Base on Fourier Analysis

A land cover classify method using remote sensing data base on Fourier Analysis was presented. For remote sensing data is discrete rather than continuous, Discrete Fourier transform(DFT) was used in temporal character analysis. With the typical vegetation phonologies´ observation data and simulated NDVI time series, multi-years AVHRR-NDVI data of six typical fields were analyzed, Form Fourier coefficients and Amplitude, three classifying factors of bias and amplitude were abstracted and used in classification. As results shown, Overall agreement between our class map and the land use map is 58.93%. The highest accuracy is forest(76.39%). And the accuracy of water, dry land and paddy land are 71.70%, 57.80% and 48.06% respectively. City are most confused by other kinds of land use, and the user accuracy is just 25.42%.