Noised Digital Number Recognition Based On BPnet

As we know, noised digital number will weaken the possibility and accuracy of recognition. Aiming this problem, noised digital number recognition based on BPnet will be discussed in this paper. Through using the high non-linear mapping from input to output of BPnet and the classifier, we have researched the designing of BPnet under noisy environment, and improvement and achievement of BPnet as well.Our research mainly focus on building a efficient classical 3 lays BP net model, which will be applied to recognition of digital number 0 to 9 then. The model should recognize the noised standard samples. There are two ways to achieve, which will be compared in this paper to have a conclusion: basic nerve net theory of number recognition; and optimized basic theory with modify factor. There is a key about the BPnet is part-minimum value, which cause the training failed when error curved surface show up. Therefore, we introduce a modify factor to optimize the BPnet, and avoid the part-minimum value. Furthermore, the system operating speed will be accelerated and the error will be less.The designing principle, process and the program of BPnet and some source program will be shown in this paper.