欢迎访问《农学学报》，
农学学报

农学学报 ›› 2025, Vol. 15 ›› Issue (12): 19-26.doi: 10.11923/j.issn.2095-4050.cjas2024-0162

• 植物保护 • 上一篇    下一篇

小麦品种及病虫害防治知识图谱构建

赵翔(), 杨婉霞(), 杨俊, 辛晨, 李奇   

  1. 甘肃农业大学机电工程学院， 兰州 730070
  • 收稿日期:2024-08-13 修回日期:2025-03-15 出版日期:2025-12-20 发布日期:2025-12-16
  • 通讯作者:
    杨婉霞，女，1979年出生，甘肃静宁人，教授，博士，研究方向：自然语言处理、知识图谱研究。通信地址：730070 甘肃省兰州市安宁区北滨河西路营门村1号 甘肃农业大学，E-mail：。
  • 作者简介:

    赵翔，男，2000年出生，四川广元人，在读硕士研究生，研究方向：自然语言处理、知识图谱研究。通信地址：730070 甘肃省兰州市安宁区北滨河西路营门村1号 甘肃农业大学，E-mail：。

  • 基金资助:
    科技创新2030“新一代人工智能”(2022ZD0115801); 智慧农场大脑知识图谱的数据采集加工及测试分析(GASU-JFSW-2023-97)

Construction of Knowledge Graph of Wheat Varieties and Pest Control

ZHAO Xiang(), YANG Wanxia(), YANG Jun, XIN Chen, LI Qi   

  1. School of Mechanical and Electrical Engineering, Gansu Agricultural University, Lanzhou 730070
  • Received:2024-08-13 Revised:2025-03-15 Online:2025-12-20 Published:2025-12-16

RichHTML

0

PDF (PC)

9

摘要：

本研究旨在解决不同地区小麦品种选育及对应病虫害防控问题。首先构建了包含3481条小麦品种数据和312条病虫害数据的数据语料库，根据语料库特征以细粒度方式定义了小麦品种与病虫害防控的知识体系模型层。其次分别在公共数据集和自建数据集上验证了Bert与Word2vec词嵌入模型在中小型数据集上的对比优势，进而采用适用于不同规模数据集的模型提取小麦品种与病虫害的属性知识。实验结果表明，在品种数据集上Bert-BiLSTM-CRF模型的F1值较Word2vec-BiLSTM-CRF模型高出0.1499；而在病虫害数据集上Word2vec-BiLSTM-CRF模型表现优于Bert-BiLSTM-CRF模型。

关键词: 知识抽取, 知识图谱, 模式层, 小麦病虫害, 小麦语料库, Bert-BiLSTM-CRF模型, 词嵌入模型

Abstract:

The paper aims to address the issues of wheat variety selection and corresponding pest and disease control in different regions. Firstly, a data corpus with 3481 wheat variety data and 312 pest data was constructed. According to the characteristics of the corpus, the knowledge system model layer of wheat variety and pest control was defined in a fine grained manner. Secondly, the comparative advantages of Bert and Word2vec word - embedding models on small and medium-sized datasets were verified on public datasets and constructed datasets respectively, and then the attribute knowledge of wheat varieties and pests was extracted by using models suitable for each scale datasets. The experimental results showed that the F1 value of Bert-BiLSTM-CRF model was 0.1499 higher than that of Word2vec-BiLSTM-CRF model in breed datasets. Word2vec-BiLSTM-CRF model was superior to Bert-BiLSTM-CRF model in pest datasets.

Key words: knowledge extraction, knowledge graph, pattern layer, wheat pests and diseases, wheat corpus, Bert-BiLSTM-CRF model, word - embedding model