基于卷积神经网络的深度学习技术在建筑现场安监场景下的应用研究——以北京市湖广会馆修缮项目为例  

Research on Application of Deep Learning Technology in Construction Site Safety Monitoring Scenarios based on Convolutional Neural Network--Taking the renovation project of Beijing Huguang Guild Hall as an example

作　　者：袁畅 杨天华 刘锋 李娇 宋晓薇 曹吉昌 王秋元 Yuan Chang;Yang Tianhua;Liu Feng;Li Jiao;Song Xiaowei;Cao Jichang;Wang Qiuyuan(Beijing Shouhua Construction and Management Co.,Ltd.,Beijing,100009;Beijing Urban Economic Society,Beijing,100028;Beijing Institute of Fashion Technology,Beijing,100029;Center of Science and Technology&Industrialization Development,Ministry of Housing and Urban-Rural Development,Beijing,100835;Lanzhou University of Technology,Lanzhou,70050)

机构地区：[1]北京首华建设经营有限公司,北京100009 [2]北京市城市经济学会,北京100028 [3]北京服装学院,北京100029 [4]住房和城乡建设部科技与产业化发展中心,北京100835 [5]兰州理工大学,兰州730050

出　　处：《建设科技》2025年第7期49-53,共5页Construction Science and Technology

摘　　要：本文以北京市湖广会馆修缮项目为案例,探讨了卷积神经网络(CNN)与深度学习技术在建筑装饰工程中的创新应用。针对传统施工管理中存在的效率低、安全隐患多等问题,研究提出了一种基于YOLO v5算法的智能监控系统,结合颜色特性化建模与目标跟踪技术,实现了施工场景中多目标的实时检测与追踪。通过构建相机RGB到CIEXYZ的颜色转换模型,并结合标准色差判据,优化了目标分类的准确性;同时,在YOLO v5s算法基础上进行模型剪枝与多线程优化,显著提升了检测速度与轻量化水平。实验结果表明,系统对安全帽颜色识别的准确率达90%以上,检测帧率(FPS)超过40,能够有效满足复杂施工场景下的实时监控需求。本研究为建筑装饰工程的智能化管理提供了技术支撑,具有较高的工程应用价值。This paper takes the renovation project of Huguang Guild Hall as case study to explore the innovative application of Convolutional Neural Networks(CNN)and deep learning technologies in building decoration engineering.To address the inefficiency and safety risks in traditional construction management,a smart monitoring system based on the YOLO v5 algorithm is proposed,which integrates color characterization modeling and target tracking technology to achieve real-time detection and tracking of multiple objects in construction scenarios.By establishing a color conversion model from camera RGB to CIEXYZ and optimizing target classification accuracy through standard color difference criteria,the system enhances reliability.Additionally,model pruning and multi-threaded optimization based on YOLO v5s significantly improve detection speed and lightweight performance.Experimental results demonstrate that the system achieves over 90%accuracy in safety helmet color recognition and a detection frame rate(FPS)exceeding 40,effectively meeting real-time monitoring demands in complex construction scenarios.This research provides technical support for intelligent management in building decoration engineering,demonstrating significant practical value.

关 键 词：卷积神经网络 深度学习 建筑装饰工程 YOLO算法 目标检测 

分 类 号：TU17[建筑科学—建筑理论]