In view of the two problems of membership function construction and reasoning rules establishment in fuzzy neural network，proposes an improved genetic algorithms to complete membership function automatically generation and the fuzzy rules automatically extraction. The use of dynamic Gaussian variation operator ensures effective search of problem space in early evolution and local search accurately in the later, improves the convergence rate and abtains streamlined and stable fuzzy neural network model. Its application to the automatic recognition of passenger cars shows the effectiveness of the method.