摘要
为提高工程设计行业人员对PID(process&instrumentation drawing)仪表的统计效率,该文提出一种识别PID的目标检测方法。在YOLO V3网络结构的基础上,进一步融合浅层与深层网络,增加一个针对小目标检测尺度;采用切片原理与随机生成技术进行数据增强,形成自建数据集PID-data;通过增加一个阈值,对软化的非极大值抑制Soft-NMS(soft non-maximum suppression)算法进行改进。实验结果显示,类别平均精准度等性能指标有明显提升,表明改进后的算法优化了YOLO V3网络结构模型,达到了短时间内识别大量PID中仪表的目的。
In order to improve the statistical efficiency of PID instrument in engineering design industry,an object detection method is proposed based on identification of PID. Based on the YOLO V3 network structure,a small target detection scale is added by further integrating shallow and deep networks. The PID-data set is formed by using the slicing principle and random generation technology to enhance the data. The soft non-maximum suppression algorithm is improved by adding a threshold value. The experimental results show that the performance indexes such as average accuracy of categories are improved significantly,indicating that the improved algorithm can optimize the YOLO V3 network structure model and achieve the purpose of identifying a large number of PID instruments in a short time.
作者
来斌
王东军
刘彦彤
王颖
LAI Bin;WANG Dong-jun;LIU Yan-tong;WANG Ying(China Petroleum Pipeline Engineering Corporation Tianjin Branch,Tianjin 300457,China)
出处
《自动化与仪表》
2023年第3期54-58,共5页
Automation & Instrumentation
关键词
PID
目标检测
YOLO
V3
数据增强
PID-data
Soft-NMS
process&instrumentation drawing(PID)
object detection
YOLO V3
data enhancement
PID-data
soft nonmaximum suppression(Soft-NMS)