基于Faster-RCNN网络的接箍自动识别方法

Automatic Identification Method of Collar Based on Faster-RCNN Network

在可视化测井中,深度对于判断油管缺陷位置、射孔位置至关重要,而现有的测深系统具有一定的深度误差。在观测井下视频人工查找接箍,存在耗时、检测速度慢等问题。先对VideoLog油气井可视化测井技术采集到的井下视频进行图像增强,引入Faster-RCNN模型,采用ResNet50网络作为特征提取网络提取接箍特征,最终通过兴趣区域池化网络和全连接层完成接箍的识别定位。该模型有着0.99的平均精度,在实验中,视频中的接箍均可被准确识别,具有识别速度快和准确率高等优点。

Depth is very important for determining the location of casing defects and perforation in visual logging.However,existing sounding systems have certain depth errors.There are some problems such as time consuming and slow detection speed when searching the collar manually by video.This paper firstly enhances the downhole video collected by VideoLog visual logging technology.The Faster-RCNN model is introduced and ResNet50 network is used as feature extraction network to extract collar features.Finally,the identification and positioning of the collar are completed through the interest area pooling network and the full connection layer.The model has an average accuracy of 0.99.In the experiment,all the connections in the video can be accurately identified,which has the advantages of fast recognition speed and high accuracy.