DOI：10.3969/j.issn.1672-3872.2026.S1.003
基金项目：南宁师范大学大学生创新创业训练计划项目（202510603607）
作 者：傅晨希1 ，黄玉妹2 ，苏宾婷1 ，李高健2 ，苏启哲2 ，彭志胜2 ，徐志刚2
（1. 南宁师范大学人工智能学院 / 曙光大数据学院，广西 南宁 530100；2. 南宁师范大学物流管理与工程学院，广西 南宁 530100）
摘 要：【目的】解决传统病虫害监测方法效率低下、主观性强、难以实时预警等问题。【方法】提出了一种基于图像识别的甘蔗病虫害早期监测与预警方法。通过构建“天-地”协同监测框架，结合农户上传的田间高清图像和遥感技术获取的宏观生长信息，形成了多尺度病虫害监测数据体系。利用卷积神经网络（CNN）对甘蔗常见病虫害进行精准图像识别，并结合历史环境数据建立病虫害发生风险预测模型，实现从静态识别到动态预警的功能升级。【结果】1）优化后的CNN模型对不同阶段病虫害的平均识别准确率达92.7%。2）多源数据融合后的病虫害区域的mIoU为0.83，能精准定位病虫害发生的核心区域。3）分级预警机制的平均预警准确率为90.5%、误报率为5.2%、漏报率为4.3%，平均响应时间为2.8 h，能够有效提高病虫害监测的准确性和及时性。【结论】本研究构建了“农户上传—云端分析—专家反馈”的闭环服务模式，为甘蔗产业的可持续发展提供了借鉴和参考。未来研究可进一步增强模型的泛化能力与鲁棒性、探索小样本学习与自监督学习策略、优化多源数据配准算法，并拓展应用至生长全周期管理、其他重要农作物的病虫害监测、产业平台的闭环联动等维度，为智能农业监测提供一套可复制、可推广的解决方案。
关键词：图像识别；病虫害；早期监测；预警系统；深度学习
Research on Early Monitoring and Warning Method of Sugarcane Diseases and Pests Based on Image Recognition
Author: Fu Chenxi1 , Huang Yumei2 , Su Binting1 , Li Gaojian2 , Su Qizhe2 , Peng Zhisheng2 , Xu Zhigang2
(1.School of Artificial Intelligence & Sugon Big Data College, Nanning Normal University, Nanning 530100, China; 2.School of Logistics Management and Engineering, Nanning Normal University, Nanning 530100, China)
Abstract: [Objective] To solve the problems of low efficiency, strong subjectivity, and difficulty in real-time warning of traditional pest and disease monitoring methods. [Method] The article proposes an early monitoring and warning method for sugarcane diseases and pests based on image recognition. By constructing a “sky-ground” collaborative monitoring framework, combined with high-definition images uploaded by farmers and macro growth information obtained through remote sensing technology, a multi-scale pest and disease monitoring data system has been formed. Utilizing Convolutional Neural Networks (CNN) for precise image recognition of common sugarcane diseases and pests, and combining historical environmental data to establish a risk prediction model for disease and pest occurrence, achieving a functional upgrade from static recognition to dynamic warning. [Result] 1) The optimized CNN model achieved an average accuracy of 92.7% in identifying pests and diseases at different stages. 2) The mIoU of the pest and disease areas after multi-source data fusion is 0.83, which can accurately locate the core areas where pests and diseases occur. 3) The average warning accuracy of the graded warning mechanism is 90.5%, the false alarm rate is 5.2%, the false alarm rate is 4.3%, and the average response time is 2.8 hours, which can effectively improve the accuracy and timeliness of pest and disease monitoring. [Conclusion] This study has constructed a closed-loop service model of “farmer upload-cloud analysis-expert feedback”, which can provide reference and guidance for the sustainable development of the sugarcane industry. Future research can further enhance the generalization ability and robustness of the model, explore small sample learning and self-supervised learning strategies, optimize multi-source data registration algorithms, and expand their applications to growth cycle management, pest and disease monitoring of other important crops, closed-loop linkage of industrial platforms, and other dimensions, providing a replicable and scalable solution for intelligent agricultural monitoring.
Keywords: image recognition; diseases and pests; early monitoring; warning system; deep learning
引文信息：[1]傅晨希，黄玉妹，苏宾婷，等.基于图像识别的甘蔗病虫害早期监测与预警方法研究[J].南方农机，2026，57(S1)：12-19.
查看全文请下载PDF文件↓
