Research On Radar Forward Modeling And Intelligent Identification Of Subgrade Shallow Disease

The existing highway subgrade diseases in China show a growing trend of variety and number,and the road safety accidents caused by subgrade diseases are more and more frequent.It is of great significance for safe operation and maintenance of highway in operation period to find out the location,shape and development degree of hidden diseases of subgrade and to deal with them pertinently.At present,nondestructive testing methods such as geological radar are more and more used in roadbed disease detection.However,with the increasing amount of detected image data,the manual data processing and interpretation method based on subjective experience judgment has been difficult to meet the growing demand of subgrade engineering maintenance work and accurate and rapid identification of disasters.In this paper,starting from the non-destructive testing of highway subgrade disease by geological radar during operation,the purpose is to realize the rapid and intelligent recognition of subgrade disease image by geological radar.Geological radar forward modeling,complex signal technology and image processing are used to study the image characteristics and signal attributes of subgrade diseases with different shapes,different development degrees and different fillings.Based on the geological radar image recognition characteristics and field detection data of subgrade disease,the geological radar subgrade disease image data set is established to realize the target detection,recognition and classification of subgrade disease image based on deep learning theory.The main research results are as follows :(1)This paper analyzes and discusses the formation mechanism of roadbed loose,void and cavity diseases.Combined with the principle and method of geological radar detection,the geological morphology and geological radar image characteristics of different roadbed diseases are summarized.(2)The multi-layer composite structure model of highway subgrade was established.The forward simulation software(Gpr Max)was used to simulate the subgrade diseases with different development degrees,different shapes,different positions and filling media.The abnormal characteristics of subgrade diseases under different conditions and the identification characteristics of subgrade disease spectrum were analyzed..(3)Based on complex signal analysis and image processing methods,the results of forward simulation of subgrade diseases are analyzed.The instantaneous attributes of amplitude,phase and frequency in the image characteristics of diseases are analyzed,and the instantaneous response characteristics of different subgrade diseases are obtained.Remove direct wave and signal gain,reduce direct wave interference,suppress strong reflection wave,gain weak reflection wave.Improve the accuracy of geological radar detection and interpretation of roadbed diseases.(4)Based on the Pytorch deep convolutional neural network framework,the object detection and recognition experiments of subgrade disease images are carried out by using SSD,YOLO v3 and YOLO v4 target detection and recognition algorithms based on regression analysis theory.The results show that the average detection accuracy of YOLO v3 is 76.69 %,the average detection accuracy of SSD is 75.07 %,and the average detection accuracy of YOLO v4 is 65.08 %.(5)Multiple evaluation indexes such as Precision,Recall,F1,Ap and Map were used to evaluate the detection efficiency and accuracy of subgrade disease images,and the reliability of YOLO v3 and SSD target detection algorithms applied to intelligent identification and classification of non-destructive detection of highway subgrade geological radar in operation period was verified.