The BP neural network is widely used for classification of remote sensing image data nowadays. But it has the usual shortcomings of multilayer sensor neural network too: the question about the number of crytic layer and the number of crytic layer node

the question about local minimum

the question about training speed

and soon. In order to solve the questions thoroughly

a sort of classification algorithm of high-rank neural network is developed in this research. This algorithm has not crytic layer

so it hasn' t the question about the number of crytic layer and the number of crytic layer node. It' s interface of model classification is nonlenear

so the question about local minimum is solved thoroughly. It' s training speed is faster and the precision of model classification is greaterthan that of the BP neural network algorithm. In this article

the structure

flow chart and course control of this algorithm is introduced detailedly. Using the hyperspectral data in the destrict of Shahe town

Beijing city

an experiment is done and a excellent result is gained. The classification precision of training sample and the classification precision of test sample are all 100 percent. It is proved that the algorithm of high-rank neural network has great advantages than other algorithms of neural network in structure

speed and precision.