基于深度学习的玉米生产过程知识图谱构建  

Construction of knowledge mapping of maize production processes based on deep learning

作　　者：彭雨侬 柳平增[1,2,3] 张艳 Peng Yunong;Liu Pingzeng;Zhang Yan(School of Information Science and Engineering,Shandong Agricultural University,Tai'an,271018,China;Key Laboratory of Intelligent Agriculture Technology of the Yellow and Huaihai Sea,Ministry of Agriculture andRural Affairs,Tai'an,271018,China;Agricultural Big Data Research Center,Shandong Agricultural University,Tai'an,271018,China)

机构地区：[1]山东农业大学信息科学与工程学院,山东泰安271018 [2]农业农村部黄淮海智慧农业技术重点试验室,山东泰安271018 [3]山东农业大学农业大数据研究中心,山东泰安271018

出　　处：《中国农机化学报》2025年第2期245-252,共8页Journal of Chinese Agricultural Mechanization

基　　金：山东省农业重大应用技术创新项目(SD2019ZZ019);山东省科技特派员项目(2020KJTPY078);山东省重大科技创新工程项目(2019JZZY010713)。

摘　　要：知识图谱是以结构化形式对知识进行存储的图数据库。针对玉米种植户无法高效、快捷地获取玉米生产过程知识问题,构建一种基于深度学习的玉米生产过程知识图谱。半结构化数据来源于在农业领域专家帮助下制定的表格形式玉米高产栽培方案;非结构化数据获取则基于OCR技术将书籍转换为TXT格式的文本文档。模式层构建则根据玉米领域数据不同特征,确定玉米品种包含1类实体、12种属性,玉米病害和玉米虫害分别包含6类实体、5类关系,形成玉米种植知识实体和关系概念层。在数据层构建过程中,通过BIOES标注法,采用BERT—BiLSTM—CRF模型对非结构化数据进行实体识别。结果表明,基于BERT—BiLSTM—CRF命名实体识别模型相比LSTM、LSTM—CRF、BiLSTM—CRF,在F1值上分别提高14.31%、7.36%、3.86%;精确率、召回率和F1值分别达到89.31%、88.54%和88.92%。构建完成的玉米生产过程知识图谱可提升用户获取玉米种植知识的效率,提高玉米种植的管理水平。Knowledge graph is a graph database that stores knowledge in a structured form.Aiming at the problem that corn growers cannot efficiently and quickly obtain the knowledge of corn production process,a knowledge graph of corn production process based on deep learning was constructed.The semi-structured data comes from the maize high-yield cultivation plan in the form of a table formulated with the help of experts in the field of agriculture;the unstructured data acquisition is based on the OCR technology to convert the book into a text document in TXT format.The schema layer was constructed based on the different characteristics of the corn field data,and it was determined that corn varieties contained 1 type of entity and 12 types of attributes,and corn diseases and corn pests contained 6 types of entities and 5 types of relationships,respectively,to form a conceptual layer of corn cultivation knowledge entities and relationships.During the construction of the data layer,the BERT—BiLSTM—CRF model was used to identify entities in unstructured data through the BIOES annotation method.The results showed that the BERT—BiLSTM—CRF named entity recognition model based on BERT—BiLSTM—CRF improved 14.31%,7.36%and 3.86%in F1 value compared to LSTM,LSTM—CRF,and BiLSTM—CRF,respectively.The precision rate,recall rate and F1 value reached 89.31%,88.54%and 88.92%,respectively.The completed knowledge graph of corn production process can improve the efficiency of users in acquiring corn planting knowledge and improve the management level of corn planting.

关 键 词：深度学习 知识图谱 玉米生产过程 玉米病虫害 玉米品种 识别模型 数据采集 

分 类 号：S513[农业科学—作物学]