Application of Neural Network Based on Simulated Annealing to Classification of Remote Sensing Image

The performance was unstable when using BP neural network to classify remote sensing images. Applying simulated annealing idea, an improved BP neural network with momentum was put forward. The improved network could self-adapt to choose momentum parameters according to annealing temperature, which was able to make the network escape from local minimum spots and converge stably. The experiments show that improved network converges more easily, its performance is steady, it has the preponderances of gradient descent with momentum and the standard BP neural network. Classification accuracy of remote sensing image is comparatively high. This method has practical application value