摘要
提出了一种面向嵌入式平台的环境监测AI模型轻量化设计与部署方案,通过模型剪枝、量化、知识蒸馏等技术结合边缘计算架构,在保证监测精度的前提下,实现模型体积压缩80%以上、推理速度加快3倍。实验结果表明,该方案在树莓派4B与NVIDIA Jetson Nano设备上可实时处理空气质量、水质和噪声污染数据,且能耗降低45%。
A lightweight design and deployment scheme for AI model of environmental monitoring oriented to embedded platform is proposed.Through model pruning,quantification,knowledge distillation and other technologies combined with edge computing architecture,the volume of the model is compressed by more than 80%and the reasoning speed is increased by three times on the premise of ensuring the monitoring accuracy.The experimental results show that this scheme can process real-time air quality,water quality,and noise pollution data on Raspberry Pi 4B and NVIDIA Jetson Nano devices,and reduce energy consumption by 45%.
作者
李佳阳
LI Jiayang(Qingdao Ecological Environment Bureau Chengyang Branch Ecological Environment Monitoring Center,Qingdao,Shandong 266109,China)
关键词
环境监测
AI模型轻量化
嵌入式部署
边缘计算
模型量化
environmental monitoring
AI model lightweighting
embedded deployment
edge computing
model quantification
