基于Faster R-CNN算法的探地雷达管线目标智能识别  

Intelligent Recognition of Pipeline Target Based on Faster R-CNN Algorithm for Ground Penetrating Radar

作　　者：胡浩帮 方宏远[1] 王复明[1] 董家修 Hu Haobang;Fang Hongyuan;Wang Fuming;Dong Jiaxiu(School of Water Conservancy Engineering,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学水利科学与工程学院,河南郑州450001

出　　处：《城市勘测》2020年第3期203-208,共6页Urban Geotechnical Investigation & Surveying

摘　　要：针对传统识别探地雷达管线目标图像时速度较慢、难以识别多个相交的双曲线特征的缺点,文章提出了一种基于区域选择的目标检测与识别算法,搭建了基于Caffe深度学习框架的图像检测试验平台。采用郑州市地下管线检测的实测探地雷达剖面图像,对Faster R-CNN模型进行对比分析和优化改进,选择试验效果较优的网络模型检测识别样本数据集。试验结果表明,模型识别的平均正确率(Average Precision,AP)可达90%以上,具有良好的识别效果与噪声鲁棒性。In view of the disadvantages of slow speed and difficulty in identifying multiple intersectant hyperbolic features in the traditional recognition for pipeline images of ground penetrating radar(GPR),this paper adressed a target detection and recognition algorithm based on region selection,and built an image detection platform based on Caffe.Based on the measured GPR profile images of underground pipeline detection in Zhengzhou,the Faster R-CNN was compared,analyzed,optimized and improved.And the network model with better experimental effect was selected to detect and identify the sample data set.The experimental results show that the AP of the model can reach more than 90%,and it has good recognition effect and noise robustness.

关 键 词：探地雷达 目标检测与识别 深度学习 FASTER R-CNN 

分 类 号：TU992[建筑科学—市政工程] TN959.3[电子电信—信号与信息处理]