基于Mask R-CNN的油田井场指针仪表识别方法研究  

Research on Mask R-CNN-based pointer instrument recognition method for oilfield wellsite

作　　者：康朝海[1] 刘杨 任伟建[1] 王树峰[2] 张永丰[2] KANG Chaohai;LIU Yang;REN Weijian;WANG Shufeng;ZHANG Yongfeng(Northeast Petroleum University Daqing,Heilongjiang Daqing 163318,China;Second Oil Production Plant of Daqing Oilfield Co.,Ltd.,Heilongjiang Daqing 163414,China)

机构地区：[1]东北石油大学电气信息工程学院,黑龙江大庆163318 [2]大庆油田有限责任公司第二采油厂,黑龙江大庆163414

出　　处：《工业仪表与自动化装置》2024年第5期76-82,107,共8页Industrial Instrumentation & Automation

摘　　要：针对无人机巡检流程中采集到井场仪表图像模糊以及油田仪表定位模型实时性较差的问题,提出一种改进后最大后验概率模型去模糊方法和基于Mask R-CNN的指针式仪表定位算法。首先,通过采用变步长LMS滤波器的方法优化图像的先验信息,根据输入数据的统计特性调整滤波器参数,生成初步的仪表图像恢复结果,从而提升了最大后验概率的去模糊效果;其次,在Mask R-CNN网络结构的基础上,选用MobileNetV3作为主干特征提取网络减少参数量,再加入注意力机制模块保证准确率以完成仪表定位。最后,实验证明,仪表图像评价指标高于其他算法,该文提出的仪表定位算法减少了48.25 M参数量,FPS值达到37.3 frame/s,准确率为94.02%。Aiming at the problems of fuzzy wellsite instrumentation images captured during the UAV inspection process and poor real-time oilfield instrumentation localization model,an improved maximum a posteriori probability model deblurring method and a pointer instrumentation localization algorithm based on Mask R-CNN are proposed.Firstly,the a priori information of the image is optimized by using a variable step size LMS filter,and the filter parameters are adjusted according to the statistical characteristics of the input data to generate the preliminary instrument image recovery results,so as to improve the de-blurring effect of the maximum a posteriori probability.Secondly,on the basis of the Mask R-CNN network structure,MobileNetV3 is selected as the main feature extraction network to reduce the number of parameters,and the attention mechanism module is added to ensure the accuracy of the Mask R-CNN network structure.Secondly,based on the Mask R-CNN network structure,MobileNetV3 is chosen as the backbone feature extraction network to reduce the number of parameters,and then the attention mechanism module is added to ensure the accuracy to complete the instrument positioning.Finally,the experiment proves that the evaluation index of instrument image is higher than other algorithms,and the instrument localization algorithm proposed in this paper reduces the number of parameters by 48.25 M,and the FPS value reaches 37.3 frames/s,with an accuracy rate of 94.02%.

关 键 词：计算机视觉 Mask RCNN MobileNetV3 仪表识别 图像去模糊 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]