Research On Automatic Recognition Method Of Ground-Penetrating Radar Images Of Internal Road Pavement Diseases Based On Deep Learning

As the most basic and extensive transportation infrastructure,highway is the main support for connecting other transportation modes and playing the overall efficiency of comprehensive transportation network,and has an irreplaceable role in the transportation system Due to the influence of man-made or natural factors,various diseases such as debonding,hollowing,loosening and cracking often appear inside the road surface,which seriously affect the operational safety and service life of the road.Ground Penetrating Radar,as a mature and advanced non-destructive testing method,combined with deep learning technology,can achieve rapid,non-destructive and intelligent detection of road pavement internal diseases,effectively improve the efficiency and accuracy of road inspection and maintenance,and ensure the long-term safety of the road,which is of key practical significance and engineering value to ensure smooth traffic flow,ensure driving safety and improve transportation efficiency.Ground Penetrating Radar(GPR),as a non-destructive,efficient and convenient detection technology,has been widely used in the detection of internal road pavement diseases.However,it is difficult to further promote the application of ground Penetrating Radar detection technology in the field of highway inspection due to the wide variety of internal road pavement diseases,the huge amount of data to be detected and the complex detection environment.At present,the main problems are:manual identification is inefficient and error is large,while the traditional machine learning method needs to rely on manual design features and personal experience,which is difficult to achieve the purpose of automatic disease identification.Many pipelines and structures exist underground,but for these diseased environments are very complicated,we lack some a priori knowledge provided by physical models,and the interpretation of underground diseases lacks standards,and the interpretation results mainly rely on personal experience judgments.When groundpenetrating radar is used for detection,due to the complexity of the environment,the detected signal may be affected by a large amount of interference and noise,and even the target signal may be completely covered,thus affecting the accurate judgment of the disease target.In view of the above problems,this study has conducted an in-depth discussion on two aspects of road subsurface disease characteristics and intelligent detection methods,and the main work includes:(1)analysis of common types of urban road underground space disease and causes,understanding the working principle of ground-penetrating radar,to provide a theoretical basis for orthorectified simulation.(2)Using GprMax software to simulate four typical diseases,such as debris,voids,cracks and loosening,and to study their characteristics and patterns on radar reflection waves and grayscale maps.The radar grey-scale images from the orthorectified simulations were observed and the patterns of the black and white bands in the images were used to summarise the characteristic patterns of the various diseases.These features provide a key reference and basis for sample labelling for subsequent intelligent disease detection studies.(3)In order to accomplish the task of real-time disease detection inside the road,this paper first studies the YOLOv5 network,combining appropriate initial anchor,CIoU loss function,etc.to improve the accuracy of the network model and accelerate the convergence speed.Since the YOLOv5s network model is complex and has defects in the feature extraction of images,this paper tries to adopt the ShuffleNetV2 module based on the YOLOv5s model to improve the lightness of the model to improve the recognition speed of the model while ensuring the high accuracy of the model.(4)In order to avoid overfitting phenomenon and enhance the generalization ability of the model,the data set is expanded by using data enhancement method.And the groundpenetrating radar is used to collect the actual road and get the ground-penetrating radar image of the actual road.The simulated ground-penetrating radar images of road internal diseases are used to form the dataset,trained on the pre-trained neural network and fine-tuned,and the trained network model is automatically recognized for the test set and the groundpenetrating radar images of the real road respectively.The experimental results show that the proposed method has a good detection and recognition effect,which confirms its effectiveness and enables the rapid detection of diseases in ground-penetrating radar images.