Remote Sensing Image Classification Based on Improved BP Neural Network

Considering the disadvantages of basic BP such as its low training velocity, the difficulty in convergence and the tendency to partial minimum value, in this paper we use MATLAB 7.0 as the experimental platform, and discuss the application of improved BP neural network which is based on the combination of principal component analysis and L-M algorithm in the classification of remote sensing images. We also use an adaptive method in the setting of tentative parameters in L-M algorithm. The experimental result shows that compared with the basic BP neural network and the Maximum Likelihood in the aspects of the convergent velocity and classification accuracy, our method has an obvious advantage over the basic BP in time consumption and classification accuracy and is comparable to the Maximum Likelihood in classification accuracy. Therefore It is an effective improvement on the BP network.