基于链置换神经网络的印刷体汉字数字识别  

Recognition of printed Chinese numerals based on DNA strand displacement neural network

作　　者：吴琦琨 吕姝颖 石晓龙 WU Qi-kun;LYU Shu-ying;SHI Xiao-long(Institute of Computer Science and Technology,Guangzhou University,Guangzhou 510006,China;School of Aerospace Engineering,Beijing Institute of Technology,Beijing 102400,China)

机构地区：[1]广州大学计算机科技研究院,广东广州510006 [2]北京理工大学宇航学院,北京102400

出　　处：《广州大学学报（自然科学版）》2023年第1期1-8,共8页Journal of Guangzhou University:Natural Science Edition

基　　金：国家自然科学基金资助项目(61772376,62072129);国家重点研发计划资助项目(2019YFA076402)。

摘　　要：链置换技术是一种体外恒温无酶的分子计算技术,近年来已成为DNA计算领域的常用技术,而人工神经网络是一种模仿生物神经网络结构和功能的计算模型。基于链置换技术可以用生物分子构建神经网络,并作为分类器用于执行各种模式识别任务。文章以链置换逻辑门为基础,构建了一个赢家通吃的分子神经网络计算系统,完成了印刷体汉字数字的模式识别任务。首先将代表数字模式的图片转化成用DNA序列编码的分子数据,再将人工合成的DNA数据链输入到分子神经网络计算系统中,该网络能够利用DNA链置换技术执行生物分子计算,从而实现对输入DNA数据模式的分类,最终的分类结果将会通过荧光分子修饰单链DNA输出,并通过光电信号转换自动识别。仿真实验和生物实验证明了基于链置换的分子神经网络可以出色地完成印刷体汉字数字识别的任务。DNA Strand Displacement(DSD)is an in vitro thermostatic enzyme-free DNA amplification technique,which has become a common technique in the field of DNA computing in recent years.Artificial neural network is a computational model that mimics the structure and function of biological neural networks.Based on DNA strand replacement technique,artificial neural networks can be constructed as classifiers for performing various pattern recognition tasks.In this paper,a winner-take-all(WTA)neural network is constructed based on DNA strand replacement reaction to accomplish the recognition task of printed Chinese numerals.First,the pictures representing digital patterns are transformed into molecular patterns represented by DNA strands,and then they are fed into the constructed DNA neural network.This network can perform biocomputations using DNA strand displacement technology to classify input patterns,and the final classification results will be represented by the output of single-stranded DNA with fluorescent molecules.Through simulation and biological experiments,we demonstrate that a winner-take-all neural network based on DNA strand replacement can perform the task of recognition of printed Chinese numerals with excellent performance.

关 键 词：DNA链置换 神经网络 模式识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]