文献类型：专利

英文题名：BP neural network based handwriting recognition system dynamic sample selecting method, involves updating network weight value by using total gradient, and identifying unknown font sample by using neural network

作者：LI D;FENG W;WANG Z;FAN Q;CAO Z

机构：[1]UNIV EAST CHINA SCI & TECHNOLOGY

申请号：CN106022273-A

申请日：2016-05-24

公开日：2016-10-12

语种：英文

收录：DERWENT

摘要：NOVELTY - The method involves determining a structure of a neural network according to characteristics of a font. A gradient descent process is processed for obtaining network weight value, learning step length, charge factor, minimum stopping error and maximum iteration times in a first iteration for calculating for a total gradient. Network weight value is updated by using the total gradient. Another iteration of a sample is selected and processed until a minimum stopping error or a maximum iteration time reached. An unknown font sample is identified by using the neural network. USE - BP neural network based handwriting recognition system dynamic sample selecting method. ADVANTAGE - The method enables achieving a gradual training sample by dynamically selecting a sample by using distance from a decision boundary and reducing a BP network training time period in an effective manner. DESCRIPTION OF DRAWING(S) - The drawing shows a block diagram of a BP neural network based handwriting recognition system. '(Drawing includes non-English language text)'