Study on Classification for Remote Sensing Image based on BP Neural Network

In order to eliminate the ambiguity and uncertainty exist in the conventional classification for remote sensing images, the BP neural network was presented. However, the BP network itself also exist some limitations and shortages which are primarily represented in the aspects of network training speed low, optimization for convergence to integer not easy and so on. This paper improves the BP neural network based on Mat Lab software by using momentum and Adaptive learning rate. After 300 times of training for a sheet of panchromatic remote sensing image, the characteristics of original image can be emulation ally output reality. The total accuracy for classification is 86.57%, Kappa coefficient is 0.82, so that the precision can meet the needs of the classification of remote sensing images.