Skip to Main Content
Neural network has been increasingly used in remote sensing image classification in the last few decades. Nevertheless, an efficient method for high resolution remote sensing image classification, particularly panchromatic (PAN) image, is still under investigation. This work presents a neural network classification method for urban land cover mapping using the wavelet-based features extracted from a PAN IKONOS image. A structured-based neural network with back propagation through structure (BPTS) algorithm is conducted for image classification. After wavelet decomposition, the object's contents from the PAN image can be represented by its wavelet coefficients. The pixels' spectral intensity and the derived wavelet coefficients are combined as attributes for the tree representation in the neural network. With the designed neural network structure, a total of 2510 pixels of four land cover classes are selected as training data and 19498 pixels for the same land cover classes are selected for testing data. All the land cover classes are perfectly classified (100%) using the selected training data and the classification rate based on testing data set reaches to 99.68%. The experimental results reveal that the proposed method demonstrates a viable solution for classification of high resolution panchromatic remote sensing data.