A New Way for Handwritten Numeral Recognition

Abstract:

Handwritten numeral recognition method generally uses neural networks, the more prominent of these is BP neural network, but BP algorithm is easily get in a local minimum of the error-prone and causes slow oscillation and training , general solution for it is to optimize the structure of the algorithms first. Therefore, on the basis of the analysis of GA-BP algorithm, propose a method of making the appropriate operators of GA such as crossover and mutation probability, optimizing the weights and thresholds of BP Neural Network with the improved GA. At handwritten numeral recognition experiment, the results show that the method has faster convergence and more reliable stability, greatly improved BP neural network for learning and recognition rate.

• Periodical: Advanced Materials Research (Volumes 798-799)
• Pages: 643-646
• DOI: https://doi.org/10.4028/www.scientific.net/AMR.798-799.643
• Online since: September 2013
• Authors: Bao Lin Guan, Li Deng Ba
• Keywords: AGA, BP, Handwritten Numeral Recognition

References

[1] C.Z. He, F.J. Li and H. Y. Practice and cognition of mathematics, Vol. 32 (2002) No. 4, p.555.

[2] X.Y. Men, X.J. Han and S Men. Chinese Control Conference(CCC) (Harbin, 2006).

[3] S.L. Wang, S, Yin and M.H. Jiang. Natural Computation(ICNC) (Jinan, 2008).

[4] Z. Michalewicz. Genetic Algorithms Data structures Evolution programs. Berlin: Springer Verlag, (1996).

[5] J.S. Kong, W.W. Zhang. Computer Simulation, Vol. 28(2011)No. 2, pp.1-3.

[6] L.H. Zhong, W. Hu. Sichuan University, Vol. 44(2007)No. 5, pp.1-4.

[7] J. Zhang, J. Cao and S. Z Jiang. Neural Network Practical Guide(Mechanical Industry Press, China. 2008, p.31 ~ 37. ).