Spatial landcover classification using a neural network driven by co-occurrence matrix for landcover elements

A new land cover classification method is proposed which utilizes a three-layered feed forward neural network as the classifier and a co-occurrence matrix measured for land cover components as a feature. The land cover components are extracted by using a clustering method. As the result, the proposed method showed the best mean classification accuracy when 128 land cover components are used. And the classification accuracy is 3% higher than that of the conventional neural network classifier based on a co-occurrence matrix for gray levels