Luo Huijuan, Liu Wenjing, Xu Pinghu, Zhang Lijun, Li Lin
National Center for Materials Service Safety, University of Science and Technology Beijing, Beijing, 100083, China.
Research Institute of Macro-Safety Science, University of Science and Technology Beijing, Beijing, 100083, China.
Heliyon. 2024 Aug 13;10(16):e36264. doi: 10.1016/j.heliyon.2024.e36264. eCollection 2024 Aug 30.
In the university laboratory environment, it is not uncommon for individual laboratory personnel to be inadequately aware of laboratory safety standards and to fail to wear protective equipment (helmets, goggles, masks) in accordance with the prescribed norms. Manual inspection is costly and prone to leakage, and there is an urgent need to develop an efficient and intelligent detection technology. Video surveillance of laboratory protective equipment reveals that these items possess the characteristics of small targets. In light of this, a laboratory protective equipment recognition method based on the improved YOLOv7 algorithm is proposed. The Global Attention Mechanism (GAM) is introduced into the Efficient Layer Aggregation Network (ELAN) structure to construct an ELAN-G module that takes both global and local features into account. The Normalized Gaussian Wasserstein Distance (NWD) metric is introduced to replace the Complete Intersection over Union (CIoU), which improves the network's ability to detect small targets of protective equipment under experimental complex scenarios. In order to evaluate the robustness of the studied algorithm and to address the current lack of personal protective Equipment (PPE) datasets, a laboratory protective equipment dataset was constructed based on multidimensionality for the detection experiments of the algorithm. The experimental results demonstrated that the improved model achieved a mAP value of 84.2 %, representing a 2.3 % improvement compared to the original model, a 5 % improvement in the detection rate, and a 2 % improvement in the Micro-F1 score. In comparison to the prevailing algorithms, the accuracy of the studied algorithm has been markedly enhanced. The approach addresses the challenge of the challenging detection of small targets of protective equipment in complex scenarios in laboratories, and plays a pivotal role in perfecting laboratory safety management system.
在大学实验室环境中，个别实验室人员对实验室安全标准认识不足，未按规定规范佩戴防护装备（头盔、护目镜、口罩）的情况并不少见。人工检查成本高且容易出现疏漏，迫切需要开发一种高效智能的检测技术。对实验室防护装备的视频监控显示，这些物品具有小目标的特点。鉴于此，提出了一种基于改进YOLOv7算法的实验室防护装备识别方法。将全局注意力机制（GAM）引入高效层聚合网络（ELAN）结构，构建兼顾全局和局部特征的ELAN-G模块。引入归一化高斯瓦瑟斯坦距离（NWD）度量来替代交并比（CIoU），提高了网络在实验复杂场景下检测防护装备小目标的能力。为了评估所研究算法的鲁棒性，并解决当前个人防护装备（PPE）数据集的不足问题，基于多维度构建了一个实验室防护装备数据集用于算法的检测实验。实验结果表明，改进后的模型平均精度均值（mAP）值达到84.2%，与原模型相比提高了2.3%，检测率提高了5%，微F1分数提高了2%。与主流算法相比，所研究算法的准确率有了显著提高。该方法解决了实验室复杂场景下防护装备小目标检测的难题，对完善实验室安全管理体系具有关键作用。
