A novel automatic classification method based on multi-spectrum remote sensing imagery

The classification of multi-spectrum remote sensing imagery is studied in this paper. Spectral information of each band is represented by quantized gray values. Considering the spatial dependence among pixels, spatial information is extracted as significant supplement to denote imagery pixel. The difference between adjacent pixels is adopted to describe the variety of pixel gray values. The gradient is introduced to detect change information in horizontal, vertical and diagonal direction. The back-propagation network is adopted to perform classification. An urban and rural connecting area of Qingdao was selected as the test region. Its multi-spectrum remote sensing imagery was acquired by SPOT5. There are five primary land cover types in the region including road, greenland, cropland, building and water. The results show that the proposed method can perform land cover classification effectively for the multi-spectrum imagery.