Neural network based remote sensing image classification in urban area

Remote sensing image classification plays an important role in a variety of urban studies. This study presents a method for panchromatic (PAN) image classification using wavelet features in neural network. A structured-based neural network with backpropagation through structure (BPTS) algorithm is conducted for PAN image classification. After wavelet decomposition, an object's contents can be reflected by its wavelet coefficients. Therefore, a pixel's spectral intensity and its wavelet coefficients can be combined as attributes for the neural network. The nodes of tree representation of an object can be represented by the attributes. In order to prove the efficacy of the proposed method, the conventional features are used in the experiments as well. 2510 pixels for four classes are randomly selected as the data set for training the neural network and 19498 pixels are selected for testing. The four land cover classes are perfectly classified using the training data. The classification rate based on testing data set reaches 99.68% which is improved by around 10% compared to the rate by conventional feature set. Experimental results show that the proposed approach for PAN image classification is much more effective and reliable.