Research on Land Cover Classification for SPOT5

The type of land cover is principally determined by spectral reflectance of land. Different surface features have distinct absorption and reflection properties on different bands. The characteristic is used to identify land cover type from multi-spectral images. Reflectance coefficients and spectral differences are extracted as feature vector to represent each pixel. Self-Organizing Feature Map neural network is adopted for feature classification. A semi-supervised correction strategy is proposed to amend the classification errors. The images of Qingdao Agricultural University campus acquired by SPOTS were selected as experimental data. The results show that the overall classification accuracy of samples is up to 83%, moreover, the proposed method is helpful in solving the phenomena of same object with different spectrums or different objects with same spectrum.