| 摘 要: 针对智慧农业数据采集的可靠性难题,本文提出一种专注于异常检测与数据恢复的“节点-边缘-云”三层架构。其中,节点端负责基础采集与预处理；边缘端利用机器学习与插值算法实时识别并修复离群点；云端则依托微调后的多模态大模型进一步提升检测与恢复精度。实验结果表明,该架构在准确率、误报率及归一化均方根误差等关键指标上均表现优异,显著增强了系统的可靠性与稳定性。 |
| 关键词: 智慧农业 异常检测 数据恢复 机器学习 多模态大模型 微调 |
|
中图分类号:
文献标识码:
|
| 基金项目: 国家自然科学基金青年科学(62003307) |
|
| Research on Anomaly Detection and Recovery Architecture for Smart Agriculture Scenarios |
|
wangrui, qiuying
|
School of Information Science and Engineering, Zhejiang Sci-Tech University
|
| Abstract: To address the reliability challenges of data collection in smart agriculture, this paper proposes a three-layer "node-edge-cloud" architecture focused on anomaly detection and data recovery. The node layer handles basic data acquisition and preprocessing; the edge layer utilizes machine learning and interpolation algorithms to identify and repair outliers in real time; and the cloud layer leverages a finely tuned multimodal large-scale model to further enhance detection and recovery accuracy. Experimental results demonstrate that this architecture exhibits superior performance in key metrics such as accuracy, false alarm rate, and normalized root mean square error, significantly enhancing the system"s reliability and stability. |
| Keywords: smart agriculture anomaly detection data recovery machine learning MLLM fine-tuning |
