Remote sensing image classification based on evidence theory and neural networks

An effective method based on evidence theory and neural networks to classify the remote sensing images is brought forward in the paper. Firstly, with the spatial information in consideration, the original image is smoothed with a modified gradient inverse weighting smoothing method, then the classification of the original and smoothed images is performed separately using a BP neural network. Finally, result comes out after fusing the two classification results with D-S evidence theory. Experimental results demonstrate that the proposed method is effective and can improve the classification accuracy.