Land-use and land-cover analysis with remote sensing images

Remote sensing is an important tool in land-use and land-cover classification. The spectrum information is the basis of remote sensing image classification. However, it is difficult to achieve accurate classification with a single feature. Spectrum and texture features are extracted in the paper. Wavelet transform is performed on multi-spectral images and approximate coefficient, horizontal, vertical and diagonal direction decomposition coefficient matrices are obtained. The decomposition coefficient matrices are reconstructed and reconstructed coefficient matrices are used to describe texture for multi-spectral remote sensing images. Spectrum feature is represented by gray values in multi-spectral bands. Artificial neural network is adopted for classification. Experimental region is a part suburban area in Qingdao. There are four land-use and land-cover types in the region, including green land, road, construction and unused land. The experimental results show that the classification accuracy is satisfactory especially in green land and construction classification.