Improved BP Neural Network Classifier Based on Decision Attribute Support and PSO and Its Application on Vehicle Classification

In this paper an improved BP neural network classifier based on decision attribute support and PSO is proposed to solve the problems of BP neural network, such as low speed in convergence, easily converging to local minimum, lacking of data analysis and reasonable guidance on the network structure design. According to the change on decision attribute support as increasing and decreasing of attributes, the minimum reduction of attributes for each classification decision is obtained. The minimum reduction of attributes is used to guide the BPNN structure design. And the structure of BPNN is optimized. The traditional gradient method is replaced with PSO global optimizing algorithm that implements global convergence of BPNN. Furthermore, this paper introduces the theory of the vehicle classification, and the improved BP neural network classifier is applied into the classification. After stimulated with actual data, the improved BP neural network classifier has been proved better than the traditional BP neural network distinctly by comparing with test result