大田作物病害识别研究图像数据集  

An image dataset for field crop disease identification

作　　者：陈雷[1] 袁媛 Chen Lei;Yuan Yuan(Institute of Intelligent Machines,Chinese Academy of Sciences,Hefei 230031,P.R.China)

机构地区：[1]中国科学院合肥智能机械研究所,合肥230031

出　　处：《中国科学数据（中英文网络版）》2019年第4期81-87,共7页China Scientific Data

基　　金：中国科学院信息化专项(XXH13505-03-104);国家自然科学基金面上项目(31871521)

摘　　要：根据联合国粮农组织报告,每年农业病虫害造成的自然损失率超过37%,农业病虫害识别与防治对于提高农业产量具有重要意义。传统人工识别方法依赖经验,主观因素较大,不够准确。近年来计算机视觉方法逐渐发展,该方法更加客观,并支持实时在线诊断,但需要大规模训练样本的支持。因此,构建可供机器学习建模使用的图像数据集对于实现高效的农业病虫害识别至关重要。为此我们构建了农业病虫害研究图库(IDADP),涵盖农业病虫害图像采集、分类、标记、存储与建模等多方面的内容,面向科研学者与农技人员两大类用户群体提供农业病害在线诊断及相关的技术咨询等服务。本数据集目前包括以水稻、小麦、玉米为主的大田作物的高质量农业病害图像数据约200 GB。与现有大多仅含有3–5幅典型症状图像的农业病害图谱类资源存在本质区别,本图像数据集由高分辨率和高相似度的同类农作物病害原始图像数据构成,每种病害的图像数量有几百乃至上千幅,可作为病害识别建模的训练样本使用。本数据集将为农业病害识别研究领域提供宝贵的基础数据资源,同时可作为大数据环境下机器学习建模的标准图库,对促进农业病害图像识别研究的发展具有重要的实际应用价值。According to the report of Food and Agriculture Organization of the United Nations,the annual natural loss rate caused by agricultural pests and diseases reached more than 37%.Identification and control of agricultural pests and diseases is significant for improving agricultural yield.Traditional manual recognition methods are not accurate enough since they rely on subjective experience.In recent years,computer vision-based methods have developed gradually.These methods are more objective and support real-time online diagnosis.As these methods depend on large-scale training samples,building an image dataset for machine learning modeling is very important for efficiently identifying agricultural diseases and pests.Therefore,we have constructed an image dataset for agricultural diseases and pests research(IDADP)which covers such aspects of agricultural diseases and pests as image acquisition,classification,labeling,storage and modeling.Meanwhile,this image dataset provides online diagnosis of agricultural diseases and related technical consultation services for scholars and agricultural technicians.The image dataset currently has about 200 GB of high-quality agricultural disease images,including field crops such as rice,wheat and corn.Essentially different from existing agricultural disease map resources which mostly contain only 3 to 5 typical symptom images,our dataset consists of the original image data of the same kind of crop diseases with high resolution and high similarity.Each disease has hundreds or even thousands of images,which can be used as training samples for machine learning modeling of disease identification.As a standard dataset for machine learning modeling in large data environment,this image dataset will provide valuable basic data resources.And it has important applicability in promoting the development of agricultural disease identification.

关 键 词：农业病害 大田作物 病害识别 标准图库 训练样本 

分 类 号：S432[农业科学—植物病理学] TP391.41[农业科学—农业昆虫与害虫防治]