结合注意力机制和编码器—解码器架构的化学结构识别方法

目的 化学结构识别是化学和计算机视觉领域的一个重要问题,传统光学化学结构识别技术在复杂化学结构识别任务中易发生信息丢失或误识别的现象,同时又因为化学物质的结构多样性常导致其无法解析,识别效果不佳。而基于深度学习的模型通常具有网络结构复杂度高、上下文信息易丢失和识别率低的问题。为此,提出一种结合注意力机制和编码器—解码器架构的化学结构识别方法。方法 首先,使用改进的ResNet50(residual network)作为特征提取器抓取表征信息;其次,使用BLSTM(bi-directional long-short term memory)作为行编码器为ResNet50提取的表征信息加强空间信息;最后,使用去填充模块和基于覆盖注意力机制的LSTM(long short-term memory)网络作为模型解码器,对化学结构图像进行解码,将编码结果解码为SMILES(simplified molecular input line entry system)序列。结果 在Indigo、ChemDraw、CLEF(Conference and Labs of the Evaluation Forum)、JPO(Japanese Patent Office)、UOB(University of Birmingham)、USPTO(United States Patent and Trademark Office)、Staker、ACS(American Chemistry Society)、CASIA-CSDB(Institute of Automation of Chinese Academy of Sciences—Chemical Structure Database)和Mini CASIA-CSDB数据集上,所提方法识别准确率分别为71.1%、70.21%、45.8%、30.3%、53.02%、58.21%、43.39%、46.3%、84.42%和85.78%,高于SwimOCSR、Image2Mol和ChemPix模型得分。结论 与其他模型相比,本文方法通过少量训练集能够获得较高的识别准确率。

Guidance and evaluation system for smooth operation of heavy-haul trains

The operation rules and methods for heavy-haul trains were studied and summarized according to the characteristics of the Daqin Railway,such as a large traffic volume,a high density and high-speed and difficult-to-operate heavy-haul trains.Combined with traction calculation and operation experience,these can be quantificationally decomposed into an evaluation standard for the smooth modularized operation of heavy-haul trains that can be recognized by computers.A train operation guidance system was designed to collect locomotive drivers’operation data,display the actual operation and standard curves in real time and give voice prompts and violation-operation alarms for safety-critical operation.In addition,software for operation analysis and evaluation was developed according to the quantified smooth operation standard.The smooth operation of heavy-haul trains was evaluated and statistically analysed through a comparative analysis of the actual operation records.Moreover,a train impact force detection device capable of monitoring the three-dimensional impact force of heavy-haul trains in real time was developed.Meanwhile,the evaluation standard for smooth operation was verified and optimized by real-time monitoring of the impact force of heavy-haul trains.Finally,on the basis of the above studies,a complete closed-loop management scheme for the smooth operation of heavy-haul trains was constructed,and the objectives of optimizing train operation strategy,standardizing drivers’operations and ensuring the smooth operation of trains were realized through application.