Research On Intelligent Recognition And Signal Processing Of Ground Penetrating Radar Images Of Highway Structural Diseases

With the rapid development of economy,more and more highway facilities are put into operation,making the structure inspection and routine maintenance of highway an important part of ensuring the long-term security service.Due to the influence of man-made and natural factors,many structural diseases such as incompleteness,voiding,and hollowing often occur during the highway operation,seriously affecting the safety and service life of the highway.It is of practical scientific significance and engineering value to study the detection and identification methods of highway structural diseases in order to ensure the long-term security service of highways.As an efficient,non-destructive and convenient detection technology,ground penetrating radar has become a common method for highway structural disease detection.However,due to the complex and diverse diseases of highway structures,the large amount of detection data and the susceptibility of radar signals to external noise,it is difficult for ground penetrating radar to be further promoted and applied in the field of highway detection.There are currently the following problems: 1)The shortcomings of manual recognition are low efficiency and large error.The traditional machine learning of disease recognition relies on personal experience and artificial design features,making it difficult to learn and extract disease features automatically,not to mention the automatic recognition of disease;2)Current highway structural disease identification is limited to "qualitative" identification,which is impossible to obtain the specific morphological characteristics of the disease and estimate the actual depth of the disease;3)In order to retain the echo information as much as possible,the ground penetrating radar often contains a large amount of noise interference when the echo signal is collected.The low noise and the effective echo signal which is completely submerged by noise interference all affect the subsequent interpretation and identification of disease targets.According to the problems mentioned above,taking the realization of automatic identification of highway structural diseases as the main goal,the paper proposed a rapid identification algorithm of highway structural diseases based on deep learning,which was used for the rapid preliminary identification of highway diseases.On the basis of rapid identification algorithm,a quantitative identification algorithm for highway structural diseases based on Mask R-CNN was also proposed for secondary quantitative identification,it could obtain the morphological characteristics of the disease target and estimate the actual depth position of the target while identifying disease.In addition,this paper proposed to use independent component analysis algorithm to deal with the noise interference in the echo signal and achieve good results.The main research work and results of this article are as follows:(1)Aiming at the problem of low efficiency of manual identification of highway structural diseases,a fast identification algorithm of highway structural diseases based on Faster R-CNN was proposed,which could survey diseases from ground penetrating radar detection images quickly.Due to the characteristics and difficulties of the highway structural disease identification,Soft-NMS and data enhancement methods were proposed to improve Faster R-CNN.At the same time,a transfer learning strategy was also introduced to effectively prevent overfitting,enhance the model robustness,and improve the identification accuracy.The superiority of the algorithm was verified by using comparative experiment with the basic version Faster R-CNN.(2)Because of the shortcomings that the existing algorithms cannot obtain the specific morphological characteristics of the disease and estimate the actual depth position of the disease,a quantitative identification algorithm of highway structural diseases based on Mask R-CNN was proposed,which could be used for the second detailed investigation of highway structural diseases.While realizing the automatic identification of highway structural diseases,the algorithm could obtain the morphological characteristics of the target from the pixel-level background and combine the Canny algorithm to extract the target contour characteristics.After curve fitting,the electromagnetic wave propagation velocity was estimated according to the target hyperbolic velocity estimation method,realizing the estimation of disease target depth position and completing the quantitative identification task at last.(3)In this paper,an independent component analysis method was proposed to process the noisy signal for the high-intensity noise interference in the echo signal.According to the characteristics of power frequency interference,the input matrix was constructed based on digital orthogonal method to solve the problem of underdetermined blind source separation.At the same time,for high-intensity random noise,a fractal dimension constrained ICA denoising method was proposed to filter out high-intensity random noise,combined with Gaussian filtering.According to analysis of simulation experiment,the proposed algorithm had good denoising performance,especially for low signal-to-noise ratio,laying foundation for the subsequent interpretation and identification of disease targets.