农业大数据学报 ›› 2025, Vol. 7 ›› Issue (3): 371-378.doi: 10.19788/j.issn.2096-6369.100060

• 数据资源 • 上一篇    下一篇

基于人工标注与对比生成模型的玉米叶病图文多模态数据集

王彦芳1,2(), 鲜国建1,2,3, 赵瑞雪1,2,3,*()   

  1. 1.中国农业科学院农业信息研究所,北京 100081
    2.国家新闻出版署农业融合出版知识挖掘与知识服务重点实验室,北京 100081
    3.农业农村部农业大数据重点实验室,北京 100081
  • 收稿日期:2025-04-16 接受日期:2025-05-20 出版日期:2025-09-26 发布日期:2025-09-28
  • 通讯作者: 赵瑞雪,E-mail: zhaoruixue@caas.cn
  • 作者简介:王彦芳,E-mail: wangyanfang01@caas.cn
  • 基金资助:
    新一代人工智能国家科技重大专项(2021ZD0113705)

Image-Text Multi-Modal Dataset of Corn Leaf Diseases based on Manual Annotation and Contrast Generation Model

WANG YanFang1,2(), XIAN GuoJian1,2,3, ZHAO RuiXue1,2,3,*()   

  1. 1. Agricultural Information Institute of CAAS, Beijing 100081, China
    2. Key Laboratory of Knowledge Mining and Knowledge Services in Agricultural Converging Publishing, National Press and Publication Administration, Beijing 100081, China
    3. Key Laboratory of Agricultural Big Data, Ministry of Agriculture and Rural Affairs, Beijing 100081, China
  • Received:2025-04-16 Accepted:2025-05-20 Published:2025-09-26 Online:2025-09-28

摘要:

玉米叶部病害的精准识别是农业智能化管理的重要环节。现有玉米病害数据集存在质量参差不齐、标签类别模糊、多模态数据匮乏等问题,尤其是中文语境下的病害描述数据的稀缺性。本研究整合了自建数据与AI Challenger、Plant Village及OpenDataLab开源的玉米叶部病害高清图像数据,并由人工基于文献、专业书籍及科学数据等先验知识对图像进行诊断性文本描述标注,共构建了中文语境下的1 653组图像-文本对多模态数据集。其中,每张图像对应的文本模态内容涵盖病害类型、病状特征及严重程度等关键信息。在此基础上,探索使用CN-CLIP与GPT2-Chinese大模型组合生成图像描述的补充增强内容,丰富描述文本模态数据的多样性,为图像自动标注提供实践验证。本数据集可为玉米病害智能诊断模型开发、中文图像描述生成及农业多模态知识图谱构建提供高质量数据样本支撑。

数据摘要:

项目 描述
数据库(集)名称 基于人工标注与对比生成模型的玉米叶病图文多模态数据集
所属学科 农业科学,计算机科学
研究主题 计算机视觉,跨模态检索,图像描述生成
数据类型与技术格式 .jpg
数据库(集)组成
数据集由图像数据和对应的文本描述数据组成,其中:图像数据集包括玉米大斑病、小斑病、褐斑病、弯孢霉叶斑病、普通锈病、南方锈病、灰斑病、圆斑病和矮花叶病等9种典型叶部病害原始图像数据,共1 653幅;文本数据,描述对应图像的作物名称、病害类型、病斑位置、数量、颜色及形状等细粒度病害特征,平均长度约32字符数,共1653条。
数据量 3.87 GB
主要数据指标 图像与其对应的描述文本
数据可用性
CSTR:17058.11.sciencedb.agriculture.00226; https://cstr.cn/17058.11.sciencedb.agriculture.00226
DOI:10.57760/sciencedb.agriculture.00226; https://doi.org/10.57760/sciencedb.agriculture.00226
经费支持 新一代人工智能国家科技重大专项(2021ZD0113705)。

关键词: 玉米, 叶部病害, 多模态数据集, 图像描述, 自动标注

Abstract:

Accurately identifying corn leaf diseases is an important part of intelligent agricultural management. The existing maize disease data sets have problems such as uneven quality, fuzzy label categories, and lack of multimodal data, especially the scarcity of disease description data in the Chinese context. This data set integrates the image data of corn disease from open source platforms such as AI Challenger, PlantVillage and OpenDataLab, and complements the high-definition disease images collected in the field. A Chinese multimodal data set containing 1653 images is constructed. Each image has its corresponding diagnostic text description, covering key information such as disease type, disease characteristics and severity. At the same time, the cn-clip and CPT2 Chinese large model are combined to achieve image description generation, which provides a method for automatic annotation. This data set can provide high-quality data support for the development of an intelligent diagnosis model of corn disease, the generation of Chinese image description and the construction of an agricultural multimodal knowledge map.

Data summary:

Item Description
Dataset name Image-Text Multi-Modal Dataset of Corn Leaf Diseases based on Manual Annotation and Contrast Generation Model
Specific subject area Agricultural Science, Computer Science
Research topic Computer vision, Cross-modal retrieval, Image captioning
Data types and technical formats .jpg
Dataset structure The dataset is composed of two parts: image data of corn leaf disease and corresponding text description data, including: 1. the original image data set of leaf disease, including 9 typical disease image data, including large leaf spot, small leaf spot, brown spot, Curvularia leaf spot, common rust, southern rust, gray spot, round spot and dwarf mosaic, with a total of 1653 pieces; 2. the diagnostic text description corresponding to the image has an average length of about 32 characters, a total of 1653.
Volume of dataset 3.87 GB
Key index in dataset Image and its corresponding description text
Data accessibility CSTR:17058.11.sciencedb.agriculture.00226; https://cstr.cn/17058.11.sciencedb.agriculture.00226
DOI:10.57760/sciencedb.agriculture.00226; https://doi.org/10.57760/sciencedb.agriculture.00226
Financial support National Science and Technology Major Project(2021ZD0113705).

Key words: corn, leaf diseases, multimodal data set, image description, automatic annotation