Abstract: The identification of crop disease is related to crop yield and quality, and it is an essential part in the development of intelligent agriculture. With the rapid development of deep learning in the field of image processing, the method of identifying crop disease types from images by deep learning has gradually become the mainstream. In this paper, we mainly review the methods of crop disease recognition based on deep learning, briefly introduce deep learning and convolutional neural network, and collect some common public disease image datasets. According to the different collection environment of training sample, we summarize the progress of deep learningbased disease identification methods in recent years from two aspects of laboratory and field, point out their advantages and disadvantages of each method, and conclude that there are three main problems in this research field such as insufficient data, difficult task and complex network structure of deep learning model. On this basis, we propose that the establishment of largescale, multispecies, and multitype disease database and the design of highperformance deep learning model are important development directions in the future.

Key words: deep learning, crop disease recognition, convolutional neural network, image recognition

摘要： 农作物病害识别关乎作物的产量与质量，是智慧农业发展过程中必不可少的重要环节。随着深度学习技术在图像处理领域的飞速发展，利用深度学习从图像中识别出农作物患病类型的方法已逐渐成为主流。主要对基于深度学习的农作物病害识别方法进行综述，简要地介绍深度学习和卷积神经网络，收集一些常用的病害图像公开数据集。根据训练样本采集环境的不同，从实验室和野外两个方面概述近年来基于深度学习病害识别方法的进展，指出每种方法的优势与不足，总结出该研究领域存在数据量不足、任务难度大、深度学习模型网络结构复杂3个主要问题，并在此基础上进行展望，提出建立大规模、多种类、多类型病害数据库和设计高性能的深度学习模型是未来的重要发展方向。

关键词: 深度学习, 农作物病害识别, 卷积神经网络, 图像识别