摘要
为提高建筑工程施工现场的安全性,提出一种基于深度学习的建筑工程施工现场监控系统。首先,根据系统目标分析,采用B/S架构将系统分为数据层、业务逻辑层、用户层3层结构;然后,根据系统功能需求,将系统功能模块分为数据采集与处理、实时通信、目标检测、可视化4个功能模块,并重点对目标检测模块进行设计。采用改进YOLOv8算法作为目标检测的基本算法,引入可变形卷积模块和SK注意力机制(Selective Kernel Attention Mechanism)提高算法的目标检测精度,同时改进边框损失函数,提高目标检测的速度。最后,将改进YOLOv8算法部署到Jetson Nano嵌入式平台上,进行建筑工程施工现场目标检测。实验结果表明,改进YOLOv8算法对建筑工程施工现场监控目标的检测精确率、召回率、平均精度均值、检测速度分别为98.18%、98.13%、98.11%、32.17FPS;系统测试表明,本系统各功能模块均通过测试,可向用户实时、准确反映建筑工程施工现场情况。因此,该系统为实现建筑工程施工全过程的可视化和智能化管理提供了技术支持,有利于提升建筑工程施工安全性。
To enhance the safety of construction sites,a deep learning-based monitoring system for construction site operations was designed.Firstly,the B/S architecture was adopted to divide the system into three layers based on system objectives:data layer,business logic layer,and user layer.Then,the system modules were categorized into four components according to functional requirements:data acquisition and processing,real-time communication,target detection,and visualization.Special emphasis was placed on the design of the target detection module.By introducing deformable convolution modules and selective Kernel attention mechanisms,the detection accuracy of the YOLOv8 algorithm was improved,the bounding box loss function was optimized to enhance detection speed,leading to the proposed improved.Finally,the improved YOLOv8 algorithm was deployed on the Jetson Nano embedded platform for construction site target detection.Experiment results demonstrated that the improved YOLOv8 algorithm achieved detection precision,recall,mean average precision,and detection speed of 98.18%,98.13%,98.11%,and 32.17 FPS,respectively,for construction site monitoring targets.System testing confirmed that all functional modules passed validation,enabling real-time and accurate monitoring of construction site conditions.Consequently,this system provides technical support for the visualization and intelligent management of the entire construction process,contributing to improved safety in construction operations.
作者
李宏伟
LI Hongwei(Guoneng Suohuang Railway Development Co.,Ltd.,Cangzhou 062350,China)
出处
《国外电子测量技术》
2025年第12期110-116,共7页
Foreign Electronic Measurement Technology
关键词
深度学习
建筑工程
施工现场
监控系统
YOLOv8算法
deep learning
construction engineering
construction site
monitoring system
YOLOv8 algorithm
