Research, Multi-sensor-based Neural Network Pattern Recognition Technology

Nerve network have the ability to automatic orgnanization and automatic studying,and can adapt to find the rule,which is concealed in the sample data.The studying ability of the nerve network is different form the traditional pattern recognition,the latter depend on the knowledge about the ruler of the recognition,while it is not necessary to know the knowledge about the ruler of recognition for the nerve network,which can get the relation of samples from the data .The main job of this paper is about how to apply Nerve network to the real time recognition. I improve the back propagation arithmetic, and apply fuzzy entropy to the training and studying of nerve network, a new arithmetic based on fuzzy entropy and error is presented. This arithmetic overcome the demerit of single rule, and is better to simulate the studying function of the brain. In order to validate the rationality and the validity of the arithmetic, 1 apply this arithmetic to the processing of information of cash and the recognition of the cash. Experiments has proved that this arithmetic have the advantage of high precision and stability and meet the need of complex system for the network's no linearity and high collateral.