Land cover classification of SPOT image by local majority voting

We proposed a hierarchy scheme for the SPOT image land cover classification. In the first level, we combine the statistical classifier, maximum likelihood classification (MLC); the neural network classifier, Learning Vector Quantization (LVQ); and use a 3 × 3 window to extract second-order statistical features to classify the image. If the pixel can't reach the same label in this stage, it is processed in the second level. In the second stage, the first-order statistical features of each point in a window region are extracted. Then, the majority voting is used to label the pixel, the central point of the window, which is unclassified in the first level.