A Target Differentiation Algorithm of Multi-neural-network based on Rough Set for Mobile Robot

Aiming at the uncertainty of sonar data in the problem of sonar-based target differentiation for mobile robot, the paper firstly presents a hierarchical reduction approach to reduce a sonar data set based on rough set theory. Then on the basis of the reduction, a multi-neural network based target differentiation algorithm is designed on the fact that the effective combination of multiple neural networks can increase the ability to pattern classification and generalization of the whole system. Parallel network architecture is adopted locally to increase the differentiation accuracy. And the average weighed strategy is applied to fuse the outputs of the parallel neural network classifiers to improve the robustness of the differentiation system to noise, physically plausible failure and missing sonar data situation. The simulation experiments prove that the new method combining with the traditional target differentiation algorithm can increase the accuracy of sonar-based target differentiation for mobile robot in indoor environments