多模融合与端边云用协同的矿工岗前健康监测管理系统

Miners" Pre-shift Health Monitoring and Management System Based on Multimodal Feature Fusion and End-Edge-Cloud-Application Collaboration

摘要:

针对矿工岗前健康监管工作质效低和数据孤岛问题，设计基于多模态融合和端边云用协同的矿工岗前健康监测管理系统；系统由身份认证和生理特征数据感知终端层、智能边缘计算层、云平台服务层以及管理交互与应用展示层组成；数据感知层采用多模摄像机、RFID非接触式终端设备和电子血压仪，构建包括矿工面部生物识别信息、物理身份信息和体温、血压、心率生理健康等2大类4模态特征数据集，经过本地编排处理后上传边缘网关；边缘网关利用轻量级多模态融合健康评估模型，实时对矿工健康状态进行分级评估；云平台服务层采用私有云实现矿工健康数据存储、趋势分析与预警、边缘评估模型优化；应用展示层通过Web和APP提供矿工岗前健康融合评估和管理可视化交互界面；仿真验证表明，多模融合算法相比规则引擎准确率相对提升约0.35%、误报率相对下降约为14.19%，漏报率相对下降约0.74%；系统已部署于陕西榆林红柳林煤矿，稳定运行2年无重大故障，人脸识别成功率、OCR 识别成功率等指标达标，系统输出的正常、警告和禁止比例符合经验预期，班组长反馈系统操作便捷，能够为矿工岗前健康筛查监测提供有效、可靠的辅助决策支持。

Abstract:

Aiming at the problems of low quality and efficiency the issue of data silos in pre?shift health supervision for miners, a pre?shift health monitoring and management system based on multimodal fusion and end?edge?cloud?application collaboration is designed. The system consists of an identity authentication and physiological data acquisition terminal layer, an intelligent edge computing layer, a cloud platform service layer, and a management interaction & application presentation layer. The data acquisition layer uses multimodal cameras, RFID contactless terminals, and electronic blood pressure monitors to construct a dataset covering two major categories and four modalities: miners’ facial biometric information, physical identity information, as well as physiological health data including body temperature, blood pressure, and heart rate. After local orchestration and processing, the data is uploaded to the edge gateway. The edge gateway uses a lightweight multimodal fusion health assessment model to perform real-time graded evaluation of miners' health status. The cloud platform service layer, built on a private cloud, provides health data storage, trend analysis, early warning, and edge model optimization. The application presentation layer offers Web and App interfaces for visualized fusion assessment and management of miners’ pre?shift health. Simulation verification shows that compared with the rule engine, the multimodal fusion algorithm achieves a relative accuracy improvement of 0.35%, a relative reduction in the false positive rate of approximately 14.19%, and a relative reduction in the false negative rate of approximately 0.74%.The system has been deployed in the Hongliulin Coal Mine in Yulin, Shaanxi Province, and has been running stably for two years without major failures. Hardware performance indicators such as face recognition success rate and OCR recognition success rate meet the required standards. The proportions of normal, warning, and forbidden outputs are consistent with empirical expectations. Team leaders report that the system is easy to operate and provides effective and reliable auxiliary decision support for preshift health screening and monitoring of miners.