The traditional BP neural network algorithm is good in learning ability and fault tolerance,while its disadvantages such as slow convergence rate and easily falling into local minimum restrict its further development and application.An improved BP algorithm with self-adaptive learning rate and additional momentum factors can effectively reduce the training time,speed up the convergence rate and inhibit the possibility of falling into a local minimum.The improved algorithm is applied to the image character recognition system.The influences of model parameters on performance of BP neural network are analyzed,and the recognition results are given after a series of parameter optimization.The experimental results show that the improved BP neural network can recognize image characters with high accuracy and robustness.