Remote sensed image classification using multi-perspective neural networks

Remotely sensed imagery classification is widely used for Earth resource inventory. Due to variations of imaging conditions the signature of images and the objects on the land have no unique correspondence. This results is a great difficulty for the computer processing of remotely sensed imagery. The present authors describe a novel neural network model LEP (Learning based on Experiences and Perspectives), and its application to remote sensed image classification. Because the network properly makes use of multi-perspective data and its learning is finely tuned by experience, the classification results have been much improved