基于计算机视觉的数据机房工业仪表识别研究  

Research on industrial instrument recognition in data room based on computer vision

作　　者：蔡登江 CAI Dengjiang(Information Technology Center,China National Offshore Oil Corporation,Beijing 100010,China)

机构地区：[1]中国海洋石油集团有限公司信息技术中心,北京100010

出　　处：《工业仪表与自动化装置》2025年第2期99-104,共6页Industrial Instrumentation & Automation

摘　　要：数据机房作为保障数据安全的重要设施,其内部的各种仪器仪表对于监控数据机房的安全至关重要。然而,传统的仪表识别依赖人工操作,易出错且效率低下。为了解决数据机房中工业仪表在复杂场景下准确率低的问题,研究提出了一种结合渐进式注意力机制与目标检测网络的仪表识别算法。此外,研究还引入了幽灵模块以增强特征表达能力,提高模型的识别准确率。实验结果表明,在不同复杂场景下的仪表识别任务中该模型均表现出较高的准确率。例如,在光照不均或有遮挡的场景A中,准确率达到了92.15%,召回率为89.47%,F_(1)分数为90.78%,处理时间为34.21 ms;在背景复杂或多仪表密集的场景B中,准确率提升至93.24%,召回率为90.85%,F_(1)分数达到92.02%,处理时间为35.17 ms。因此研究所提出的基于计算机视觉的数据机房工业仪表识别模型在复杂场景下的工业仪表识别任务中取得了显著成果,显示出较高的实用价值。As an important facility for ensuring data security,the various instruments and meters inside the data center are crucial for monitoring the security of the data center.However,traditional instrument recognition relies on manual operation,which is prone to errors and inefficient.In order to solve the problem of low accuracy of industrial instruments in complex scenarios in data centers,a instrument recognition algorithm combining progressive attention mechanism and object detection network is proposed.In addition,the study also introduced ghost modules to enhance feature expression ability and improve the recognition accuracy of the model.The experimental results show that the model exhibits high accuracy in instrument recognition tasks in different complex scenarios.For example,in scene A with uneven lighting and occlusion,the accuracy reached 92.15%,the recall rate was 89.47%,the F_(1) score was 90.78%,and the processing time was 34.21 ms.In scene B with complex background and dense instruments,the accuracy is improved to 93.24%,the recall rate is 90.85%,the F_(1) score reaches 92.02%,and the processing time is 35.17 ms.Therefore,the computer vision based industrial instrument recognition model proposed by the research institute has achieved significant results in industrial instrument recognition tasks in complex scenarios,demonstrating high practical value.

关 键 词：计算机视觉 YOLOv5 仪表识别 工业机房 

分 类 号：TP998[自动化与计算机技术]