Recognition fusion based on DSmT with BP neural network

In order to improve the recognition rate under the environment with violent noise and jam, or under the condition that the data have been polluted. The authors introduce a method of recognition fusion by combining the BP neural network and the DSmT. In single sensor, after the pattern class is recognized by BP neural network, the results are transformed to DSmT generalized basic belief assignments (gbba). And then, The DSmT combination rule is adopted to fuse the outputs of all sensors. For transforming the single sensor recognition result to DSmT gbba, the authors study and present a method that by computing the Minkowski distance between the output of the waiting to be recognized pattern and the classified outputs of the samples, and then the nearnesses between them can be computed, finally the gbba of the single sensor can be gotten by normalizing the nearnesses. It is proved by experiment that the method of recognition fusion based on DSmT with BP neural network can improve the recognition rate.