in terms of problems in feature extraction of traditional image recognition algorithm and classifier design, RMF neural network is proposed to be used firstly based on the color feature and single feature of texture for initial recognition of image and based on RBF neural network output as evidence to establish basic probability assignment; then the recognition result of RBF neural network is fused according to DS evidence theory and finally the image recognition result is obtained. The simulation shows that the mean accurate rate of recognition of image for DS-RBF reaches 92.19% far more than the recognition result of single feature and the recognition result of simple combination, and it significantly improves the stability and accurate rate of the image recognition and makes the result more reliable.
展开▼
