基于机械视觉检测的安全检测与管理优化研究
Research on Safety Detection and Management Optimization Based on Machine Vision Inspection

针对传统安全检测效率低、管理粗放等问题，本文提出一种融合改进深度学习模型与闭环管理策略的机械视觉安全检测与管理优化体系。在检测层，基于YOLOv8架构，集成CBAM注意力机制、自适应锚框生成及EIoU损失函数优化，构建轻量化目标检测模型，显著提升对安全帽、防护服及危险行为等关键要素的识别精度；在管理层，结合设备全生命周期管理理念与人员行为智能分析，设计“数据采集–风险评估–工单执行–效果反馈”闭环框架。在自建工地安全数据集(12,850张图像)上的实验表明，改进模型mAP@0.5达96.2%，单帧推理时间低于25 ms；管理策略在3个月企业试点中使安全隐患平均响应时间缩短42%，设备维护合规率提升至98%。本研究实现了检测算法与管理策略的深度协同，为机械工程安全生产智能化提供了可复用的技术–管理融合范式。
Aiming at the problems of low efficiency of traditional safety detection and extensive management, this paper proposes a machine vision safety detection and management optimization system integrating improved deep learning models and closed-loop management strategies. At the detection level, based on YOLOv8 architecture, a lightweight target detection model is constructed by integrating CBAM attention mechanism, adaptive anchor box generation and EIoU loss function optimization, which significantly improves the recognition accuracy of key safety elements such as safety helmets, protective clothing and dangerous behaviors; at the management level, combined with the concept of equipment full life cycle management and intelligent analysis of personnel behavior, a closed-loop framework of “data acquisition—risk assessment—work order execution—effect feedback” is designed. Experiments on a self-built construction site safety dataset (12,850 images) show that the improved model achieves mAP@0.5 of 96.2% with single-frame inference time less than 25ms; the management strategy reduces the average response time to safety hazards by 42% and improves equipment maintenance compliance rate to 98% in a 3-month enterprise pilot. This study achieves deep collaboration between detection algorithms and management strategies, providing a reusable technology-management integration paradigm for intelligent safety production in mechanical engineering.