The Recognition of Land Cover with Remote Sensing Image Based on Improved BP Neutral Network

Land cover and land use classification with Remote Sensing (RS) image is used broadly in dynamic monitoring of land use. For the RS image classification, the method of BP neutral network with one single hidden layer has been widely used. But the traditional BP neutral network based on gradient descendent of error has low classification rate. It is not easy to converge and often get into local minimum value. In the paper, the algorithm based on Levenberg-Marquardt (L-M) is used to improve the BP neutral network and then be applied in recognition of land cover with RS image. In the recognition test, the comparison of classification precision and convergent speed between normal BP neutral network and improved BP neutral network is processed. The test proves that the improved BP neutral network based on L-M algorithm can get higher precision of classification and faster speed than normal BP neutral network in recognition of land cover with RS image.