摘要
大语言模型(LLMs)作为人工智能领域发展的核心驱动力,凭借其强大的语义理解、逻辑推理与多模态信息处理能力,实现了对海量非结构化数据的深度解析与智能生成。近年来,大语言模型技术凭借其在处理复杂农业信息、辅助精准决策方面的卓越表现,在提升知识获取效率、赋能病虫害智能诊断、优化生产管理等方面展现出巨大潜力,有望为推动智慧农业向更高阶的认知智能发展提供核心技术支撑。本文梳理了大语言模型从基础模型演进到应用生态繁荣的全景,分析了其在农业知识问答、多模态应用和决策支持系统中的研究现状,并指出了当前研究在数据稀缺、模型可靠性与部署成本等方面面临的挑战;进一步,归纳了以检索增强生成(RAG)、多模态融合和智能体(Agent)技术为核心的解决方案,并探讨了其在克服模型知识局限、融合多源数据等方面的关键作用;通过剖析病虫害智能诊断、精准作物管理等典型应用实例,阐明了大语言模型赋能现代农业的价值与意义。最后,对大语言模型在农业领域的未来发展进行了展望。其技术演进将聚焦于模型与知识图谱的深度结合、多模态感知能力的全面增强,以及基于多智能体的农业决策支持系统建立。这些进展将共同推动农业智能从被动的辅助工具向主动的决策伙伴角色转变,最终为实现农业生产精细化管理和可持续发展注入强大的科技动能。
As a core driving force in artificial intelligence,large language models(LLMs)leverage their powerful capabilities in semantic understanding,logical reasoning,and multimodal information processing to achieve deep analysis and intelligent generation from vast unstructured data.In recent years,through their exceptional performance in processing complex agricultural information and aiding precision decision-making,LLM technologies have shown immense potential to enhance knowledge acquisition,empower intelligent pest and disease diagnosis,and optimize production management.They were poised to provide foundational technological support for elevating smart farming to a more advanced stage of cognitive intelligence.The landscape of LLMs was systematically reviewed from the evolution of foundational models to their flourishing application ecosystem.It provided an in-depth analysis of the current research status in agricultural question answering,multimodal applications,and decision support systems,while also identifying challenges such as data scarcity,model reliability,and deployment costs.Furthermore,it was outlined the key solutions centered on retrieval-augmented generation(RAG),multimodal fusion,and agent technology,discussing their critical roles in overcoming model knowledge limitations and integrating multi-source data.Through an analysis of typical application cases like intelligent pest and disease diagnosis and precision crop management,the value and significance of LLMs in empowering modern agriculture were clarified.Finally,it provided an outlook on the future development of LLMs in agriculture.It posited that future technological evolution would focus on the deep coupling of models with knowledge graphs,the comprehensive enhancement of multimodal perceptual capabilities,and the establishment of responsible AI governance frameworks.These advancements were expected to collectively drive the transformation of agricultural intelligence from a passive auxiliary tool into an active decision-making partner,ultimately providing a powerful technological impetus for achieving refined management and sustainable development in agriculture.
作者
王耀君
徐国威
朱建军
别宇辉
WANG Yaojun;XU Guowei;ZHU Jianjun;BIE Yuhui(College of Information and Electrical Engineering,China Agricultural University,Beijing 100083,China)
出处
《农业机械学报》
北大核心
2025年第9期240-256,共17页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家重点研发计划项目(2024YFD2000805、2023YFD1701000)
北京市乡村振兴农业科技项目(NY2502020025)
中央高校建设世界一流大学(学科)和特色发展引导专项资金项目(2025AC030、2025AC044)。
关键词
农业大语言模型
检索增强生成
农业知识问答系统
智能体
多模态
农业决策支持系统
large language model for agriculture
retrieval-augmented generation
agricultural knowledge question-answering system
agent
multimodal
agricultural decision support system
