基于改进YOLOv8的建筑工地智能监控系统设计与实现
首发时间:2026-03-13
摘要:针对建筑工地传统人工监管效率低、覆盖面有限等问题,本文设计并实现了一套基于改进YOLOv8的智慧工地安全监控系统,聚焦安全帽佩戴检测与火灾预警两大核心场景。在安全帽检测方面,为解决小目标检测难及遮挡识别率低的问题,引入了SPD-Conv空间深度转换卷积、P2小目标检测层、坐标注意力机制(CA)和WIoU v3损失函数。改进后模型mAP@0.5由86.3%提升至92.1%,小目标检测精度提升11.2%。在火灾检测方面,为降低复杂背景下的误报率并适应边缘设备的算力限制,模型采用了FasterNet轻量化主干网络,引入BiFPN增强多尺度特征融合,融合SimAM无参注意力机制,并对预测头进行了边缘特征优化。改进后参数量仅为8.5M,mAP@0.5达93.7%,误报率由28.6%降至7.9%。此外,系统通过MQTT协议对接CTWing物联网平台,实现了数据的实时上报与报警远程推送,传输成功率达97%以上。本研究为建筑工地安全管理提供了高效、可靠的智能化解决方案。
关键词: 计算机应用 深度学习 YOLOv8 目标检测 安全帽识别 火灾检测 物联网
For information in English, please click here
Design and Implementation of Intelligent Construction Site Monitoring System Based on Improved YOLOv8
Abstract:Addressing the inefficiencies of traditional manual supervision on construction sites, this paper proposes an intelligent safety monitoring system based on an improved YOLOv8, focusing on safety helmet detection and fire warning. For helmet detection, the model integrates SPD-Conv, a P2 layer, Coordinate Attention (CA), and WIoU v3 to tackle small object and occlusion challenges. This boosts mAP@0.5 from 86.3% to 92.1% and improves small object detection by 11.2%. For fire warning on resource-constrained edge devices, the network employs a lightweight FasterNet backbone, BiFPN, SimAM, and an optimized prediction head. This reduces parameters to 8.5M while achieving a 93.7% mAP@0.5 and dropping the false alarm rate from 28.6% to 7.9%. Furthermore, the system uses MQTT to connect with the CTWing IoT platform, enabling real-time remote alarms with a >97% transmission success rate. This research provides a reliable and efficient solution for smart construction safety management.
Keywords: computer application deep learning YOLOv8 object detection helmet recognition fire detection IoT
基金:
引用

No.****
同行评议
共计0人参与
勘误表
基于改进YOLOv8的建筑工地智能监控系统设计与实现
评论
全部评论