摘要
为实现化学实验室动火实验无人值守行为的智能化识别,提出基于目标检测算法的识别方法。定义动火实验无人值守行为并设计6组场景模拟方案,通过对3种不同行为模式进行模拟,构建模型训练与验证所需图像数据集,并基于目标检测算法构建化学实验室动火实验无人值守行为识别模型。研究结果表明:融合SimAM注意力机制且使用WIoU损失函数的模型识别精确率达0.996,召回率0.996,mAP@0.5为0.996;泛化能力测试精确率为0.891,召回率为0.783,mAP@0.5为0.821,融合SimAM注意力机制且使用WIoU损失函数后的模型的识别能力和泛化能力均得到明显提升。研究结果可为化学实验室动火实验无人值守行为识别与预警提供技术支撑。
In order to realize the intelligent recognition of unattended behavior in hot experiments of chemical laboratory,a recognition method based on target detection algorithm was proposed.The unattended behavior of hot experiments was defined,and six groups of scene simulation schemes were designed.Through the simulation of three different behavior modes,the image dataset required for model training and verification was constructed,and the recognition model for the unattended behavior of hot experiments in chemical laboratory was constructed based on the target detection algorithm.The results show that the recognition precision rate,recall rate and mAP@0.5 of the model integrating SimAM attention mechanism and using the WIoU loss function are 0.996,0.996 and 0.996,respectively.In the generalization ability test,the precision rate is 0.891,the recall rate is 0.783,and mAP@0.5 is 0.821.The recognition ability and generalization ability of the model after integrating the SimAM attention mechanism and using the WIoU loss function are significantly improved.The research results can provide technical support for the recognition and early warning of unattended behavior in hot experiments of chemical laboratories.
作者
徐浩钧
胡啸峰
吴建松
XU Haojun;HU Xiaofeng;WU Jiansong(School of Information and Cyber Security,People’s Public Security University of China,Beijing 100038,China;Center for Capital Social Safety,People’s Public Security University of China,Beijing 100038,China;Institute for Emergency Rescue Ergonomics and Protection,China University of Mining and Technology(Beijing),Beijing 100083,China;School of Emergency Management and Safety Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China)
出处
《中国安全生产科学技术》
CAS
CSCD
北大核心
2023年第12期135-141,共7页
Journal of Safety Science and Technology
基金
国家自然科学基金项目(72174203)。
