Improvements to FP algorithm in feedforward neural networks and its application in noisy character recognition

In this paper, we conducted the research on the classification performance of forward propagation (FP) algorithm in feed-forward neural networks, and explored how to develop its superiority fully To deal with problems such as high reject rate, and also to enhance its classification ability and robustness in recognition applications, we made simple but effective improvements to FP algorithm in network structure, threshold calculation and recognition strategy. In this process, the idea of fuzzy classification' was formed and proposed. We built a software system utilizing the improved FP algorithm to recognize noisy characters and digits from car plates. The satisfactory experimental results verified the effectiveness of the improved algorithm.