摘要
采用随机对角Levenberg-Marquardt算法有效改进了Simard卷积网络的收敛速度,分析了样本类别数、全局学习率对网络收敛速度的影响,并成功地把Simard网络推广到对百度验证码等多字体小字符集的识别,达到98.4%的单字符识别率和93.5%的整体识别率.实验表明:改进后的Simard网络具有前期预处理少、泛化能力强、收敛速度较快的优点,可以胜任多字体小字符集的识别工作.
By using the stochastic diagonal Levenberg-Marquardt method into convolutional network presented by Simard,it was analysed the relations between sample class number,global learning rate and network′s convergence speed.Simard network was then extended to multi-fontstyle little character set such as Baidu CAPTCHA.A recognition rate 98.4% in single Baidu CAPTCHA character,and 93.5% as whole was reached.Experimental results confirmed that Simard network could be used in recogniton of mutli-fontstyle little character set.
出处
《浙江师范大学学报（自然科学版）》
CAS
2011年第4期425-428,共4页
Journal of Zhejiang Normal University:Natural Sciences
