作物病虫害多模态智能管控技术研究进展

Research Progress in Multimodal Intelligent Management and Control Technologies for Crop Pests and Diseases

摘要:

作物病虫害具有发生机理复杂、影响因素多、时空异质性强等特点，现有研究多聚焦单一模态或单项任务，难以支撑诊断、监测、预警和防控服务的协同开展。在总结国内外相关研究进展的基础上，对作物病虫害多模态智能管控关键技术进行了系统分析与讨论。梳理作物病虫害发生机理与多因素耦合特征，归纳显微图像、可见光图像、多/高光谱图像、热红外、环境气象、遥感观测、农事记录和电子病历等多模态数据类型及其预处理方法;构建点-面-时空"多尺度智能管控框架，并进一步提出作物病虫害多模态智能管控总体技术框架;围绕点尺度智能识别与精准防治、面尺度区域监测与精准作业、时空尺度动态预警与决策服务、全流程智能管控系统集成与应用示范等方面总结相关研究进展。综述分析结果表明，作物病虫害智能管控正由单一识别向多模态、多尺度协同分析发展，由传统智能模型向知识增强农业大模型与智能体发展，由方法研究向云边端协同与机器人闭环服务发展。但在数据标准化、模型泛化、平台协同、实时控制和推广落地等方面仍存在明显不足。未来应推动多源异构数据联通、知识增强建模、智能装备协同和可信闭环服务体系建设，为作物病虫害精准、高效和绿色防控提供技术支撑。"

Abstract:

Crop pests and diseases are characterized by complex occurrence mechanisms, multiple influencing factors, and strong spatiotemporal heterogeneity.Existing studies have mostly focused on single modalities or individual tasks, making it difficult to support the coordinated implementation of diagnosis, monitoring, early warning, and prevention-control services.Based on a review of domestic and international research progress, it systematically analyzed the key technologies for multimodal intelligent management and control of crop pests and diseases.Firstly, the occurrence mechanisms and multi-factor coupling characteristics of crop pests and diseases were summarized, and the multimodal data types and preprocessing methods were reviewed, including microscopic images, visible images, multispectral/hyperspectral images, thermal infrared images, environmental and meteorological data, remote sensing observations, farming records, and plant electronic medical records.Secondly, a "point-area-spatiotemporal" multi-scale intelligent management framework was constructed, and an overall technical framework for multimodal intelligent management and control of crop pests and diseases was further proposed.Finally, research progress was summarized in terms of point-scale intelligent identification and precise prevention-control, area-scale regional monitoring and precision operation, spatiotemporal dynamic early warning and decision-support services, and the integration and application demonstration of full-process intelligent management systems.It showed that intelligent management and control of crop pests and diseases was developing from single-task identification toward multimodal and multi-scale collaborative analysis, from traditional intelligent models toward knowledge-enhanced agricultural foundation models and agents, and from methodological research toward edge-cloud-end collaboration and robotic closed-loop services.However, significant challenges remained in data standardization, model generalization, platform collaboration, real-time control, and practical deployment.In the future, promoting multi-source heterogeneous data interconnection, knowledge-enhanced modeling, intelligent equipment collaboration, and trustworthy closed-loop service systems would provide technical support for precise, efficient, and green prevention and control of crop pests and diseases.