基于残差胶囊增强网络的手写化学方程式配平  

The Realization of Capsule Balance Based on Handwritten Chemical Equation

作　　者：李大舟 张小明 高巍 LI Dazhou;ZHANG Xiaoming;GAO Wei(Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学计算机科学与技术学院,辽宁沈阳110142

出　　处：《沈阳化工大学学报》2023年第1期87-96,共10页Journal of Shenyang University of Chemical Technology

基　　金：辽宁省教育厅科学研究项目(LQ2017008);辽宁省高等教育学会“十三五”规划高教研究项目(GHYB160163);沈阳化工大学教育教学培育工程项目(2020 No.35)。

摘　　要：针对人工手动录入化学方程式至电子文档效率低、手动配平化学方程式高计算量、难度大的问题,提出一种残差胶囊增强网络(residual capsule enhanced network)自动识别和自动录入手写化学方程式,并且计算机自动配平方程式的新方法.为准确将化学方程式分割为单个化学字符,采用改进的投影法和最小外边框组合算法进行化学方程式切分;在传统胶囊网络的基础上,将残差卷积子网络和增强型胶囊进行结合,并引入两个级别的初级胶囊以提高网络获取多样性特征性能,进而解决化学方程式识别率低的问题;通过采用改进的高斯消元算法解决化学方程式自动配平问题.对比实验结果表明:残差胶囊增强网络识别率为94.97%,优于传统的经典算法.改进的高斯消元配平算法的准确率为95.35%,比传统的高斯消元配平算法提升了9.46%.该算法在手写化学方程式自动识别及配平方面具有一定的应用价值.In view of the low efficiency of manual input of chemical equations into electronic documents,high computational complexity and difficulty in manual balancing chemical equations,a new method for automatic recognition and automatic input of handwritten chemical equations by residual capsule enhanced network and automatic square matching program by computer was proposed.In order to segment the chemical equation into a single chemical character accurately,the improved projection method and the minimum outer border combination algorithm are used to segment the chemical equation.On the basis of the traditional capsule network,the residual convolution sub network is combined with the enhanced capsule,and two levels of primary capsule are introduced to improve the performance of the network to obtain the diversity characteristics,and then the chemical equation recognition is solved.An improved Gaussian elimination algorithm is used to solve the problem of chemical equation automatic balancing.The experimental results show that the recognition rate of residual capsule enhanced network is 94.97%,which is better than the traditional classical algorithm.The accuracy of the improved Gaussian elimination algorithm is 95.35%,which is 9.46%higher than the traditional Gaussian elimination algorithm.This algorithm has certain application value in automatic recognition and balancing of handwritten chemical equations.

关 键 词：胶囊网络 胶囊增强算法 残差卷积 动态路由 高斯消元算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]