Research on Algorithms of Neural Network Ensemble with Multi-dsp Mixture Structure

A neural network ensemble algorithm used in automatic target recognition (ATR) was designed based on multi-DSP mixture structure. It is formed a same type classifier under different size and initial condition by BP neural networks, and map every output of neural network into posteriori probability, then make weighted summation and judge. Real-time recognition was run by the neural networks ensemble system. The result showed that statistic recognition rate was up to 97 percent for 8 type object. The processing was finished in 30 ms. The recognition rate was better than single neural networks obviously.