【摘要】 病害是影响农作物品质和产量的重要因素，随着计算机视觉、光学、遥感和物联网技术的进步，基于图像的作物病害自动识别研究发展迅速.为深入了解全球作物病害图像识别的相关研究进展，利用文献计量分析方法对Web of Science核心合集(SCI-E)2002-2022年间作物病害图像识别研究领域发表的相关文献进行分析.结果表明：作物病害图像识别研究呈明显上升趋势；学科领域涉及计算机科学、农学、植物科学、工程、环境生态学、遥感等，体现出明显的综合性和交叉性特点；中国、美国、印度、德国等国家发文数量最多，整体而言各国之间均存在较为密切的交流与合作，其中中美之间合作最为密切；在发文量排在世界前10的研究机构中有6家来自中国，展现出很强的整体优势；MAHLEIN A K、 HUANG W J和KHAN M A是发文量排在前3的核心作者；Computers and Electronics in Agriculture、Frontiers in Plant Science、Remote Sensing等期刊为主要发表载体；作物病害图像数据的获取、基于机器学习的作物病害图像识别以及基于深度学习的作物病害图像识别是近20年该研究领域的主要热点和重点.作物病害图像识别的研究深受先进技术推动，尤其是在当前人工智能技术背景下方兴未艾，是面向智慧农业的重要组成部分.而当前数据样本规模偏小，相似症状的不同病害精确识别困难，模型可解释性和泛化性有限等问题依旧制约其进一步发展.构建基于生成式大模型的大规模作物病害数据集，加强多模态数据融合，提升模型的可解释性和泛化性，开展实时监测识别等内容将是未来作物病害图像识别的主要研究方向.更多还原

【Abstract】 The threat of crop diseases is one of the most important factors affecting crop quality and yield. With the advance of computer vision, optics, remoting sensing, and the Internet of Things(IoT), research on imaged-based automated identification and diagnosis of crop diseases has grown rapidly. In order to gain an in-depth understanding of the global research progress in crop diseases image recognition, this study utilized bibliometric analysis methods to analyze the relevant literatures indexed by the Web of Science Core Collection(SCI-E) database from 2002 to 2022. It is found that crop diseases image recognition research has shown a significant upward trend, and the subject areas include computer science, agronomy, plant science, engineering, environmental ecology, and remote sensing, etc., demonstrating substantial interdisciplinary and cross-disciplinary characteristics. Most publications come from China, the United States, India and Germany, and there is a close cooperation among countries, with China-US cooperation being particularly extensive. Six Chinese research institutions are ranked among the top 10 research institutions in terms of publications, showing a strong overall advantage. The top 3 core authors are MAHLEIN A K, HUANG W J, and KHAN M A. Computers and Electronics in Agriculture, Frontiers in Plant Science, Remote Sensing and other journals are the main publication carriers. The acquisition of crop diseases image data, machine learning-based crop diseases image recognition, and deep learning-based crop diseases image recognition have been the major research focuses in this filed. Technology has greatly accelerated the crop diseases image recognition research, especially in light of artificial intelligence technology, which is an important component of smart agriculture. However, the current development is limited by the lack of large-scale data sets, difficulties in identifying the diseases with similar symptoms, and weak interpretability and generalization of the diseases recognition models. The research direction of crop diseases image recognition can be expected as the construction of large-scale dataset using generative large models, the enhancement of multimodal data fusion, the improvement of interpretability and generalization of the models, as well as the conduct of real-time screening and monitoring in the future.更多还原