CN-121997244-A - Roadbed cavity intelligent detection method and relevant equipment based on multi-feature fusion and deep learning

CN121997244ACN 121997244 ACN121997244 ACN 121997244ACN-121997244-A

Abstract

The application relates to the technical field of roadbed cavity detection, in particular to a roadbed cavity intelligent detection method based on multi-feature fusion and deep learning and related equipment. The method comprises the steps of obtaining roadbed land pulse signal data and preprocessing the roadbed land pulse signal data to form a preprocessed signal sequence, extracting multidimensional features such as amplitude attenuation and dispersion change based on a cavity physical effect model, fusing the features to form fused feature vectors with high discriminant, training a deep learning cavity recognition model by utilizing the fused features to realize automatic recognition of the cavity existence probability of a road section to be detected, and finally carrying out space positioning and risk assessment by combining cross-correlation analysis and dispersion curve inversion to realize high-precision detection and reliable judgment of roadbed cavities. The method improves the accuracy, efficiency and intelligent degree of roadbed cavity detection.

Inventors

XU JIN
LI SHAOFENG
YU PENG
WU CHANGSHENG
SONG JIAN
CHEN SHUILONG
LIU XIAOTING
PENG ZHAO
YI WENHU
Tang Roumei

Assignees

广州市市政工程设计研究总院有限公司
广州市白云区重点交通项目管理中心

Dates

Publication Date: 20260508
Application Date: 20251219

Claims (10)

1. The intelligent roadbed cavity detection method based on multi-feature fusion and deep learning is characterized by comprising the following steps: obtaining ground pulse signal data of a target roadbed area; performing frequency domain filtering and self-adaptive noise suppression on the ground pulse signal data to form a preprocessing signal sequence; Based on a physical effect model of the cavity, extracting characteristics of the preprocessing signal sequence to generate a multi-dimensional characteristic set; performing feature fusion and dimension reduction on the multi-dimensional feature set to obtain a fusion feature vector; optimizing the deep learning cavity recognition model based on the fusion feature vector; inputting the preprocessed signal sequence of the road section to be tested into the optimized deep learning cavity recognition model to obtain the probability of cavity existence; based on the probability of existence of the cavity, combining the cross-correlation analysis and the dispersion curve inversion result, completing space positioning and risk assessment of the cavity, and obtaining a cavity detection result, wherein the cavity detection result comprises space position or high risk early warning information.
2. The method of claim 1, wherein the physical effect model comprises: A scattering effect sub-model for performing Gaussian attenuation modeling on scattering intensity in the pre-processing signal sequence based on the distance change between the station and the cavity center; a resonance effector model for identifying energy aggregation characteristics of specific frequency components in the pre-processed signal sequence based on physical property differences between the cavity and surrounding medium; And the wave velocity abnormal effector model characterizes low-speed deviation of the phase velocity in the preprocessing signal sequence in a target frequency band based on equivalent stiffness change caused by the cavity.
3. The method of claim 2, wherein the scattering effector model is formulated as follows: ; Wherein, the Is shown in the first The intensity of scattering effects observed at the individual stations; Representing the maximum effect amplitude at the center of the cavity; Represent the first The location of the individual stations; representing the position of the center of the cavity; Representing the standard deviation of the control effect attenuation range.
4. The method of claim 1, wherein the set of multi-dimensional features comprises: Time domain and nonlinear features for quantifying scattering effects, including root mean square, peak, skewness, kurtosis, sample entropy, and Lyapunov index; the frequency domain features are used for quantifying the resonance effect and comprise a main frequency, a bandwidth, a spectrum entropy and an H/V spectrum ratio which are obtained through fast Fourier transformation; And the time-frequency domain features are used for quantifying wave velocity abnormal effects, and comprise energy concentration area features extracted through wavelet transformation.
5. The method of claim 1, wherein the deep learning hole recognition model comprises a convolutional neural network model, a long-term memory network model, and an attention model.
6. The method of claim 5, wherein optimizing the deep-learning hole recognition model based on the fused feature vector comprises: performing multi-scale convolution extraction through the convolution neural network model based on the fusion feature vector to form local feature mapping, wherein the local feature mapping characterizes scattering anomaly, resonance frequency band change and wave velocity disturbance; inputting the local feature map into the long-short-term memory network model for time sequence association modeling to obtain a hidden state sequence reflecting the evolution rule of the cavity physical effect; Feature weighting is carried out through the attention model based on the hidden state sequence, so that a global representation vector which highlights key physical response is obtained; Constructing a feature expression layer and a parameter frame of a deep learning cavity recognition model based on the weighted feature representation; And training and converging parameters of each network structure in the deep learning cavity recognition model through the labeling sample.
7. The method of claim 1, wherein the performing spatial localization and risk assessment of the hole based on the probability of existence of the hole in combination with cross-correlation analysis and dispersion curve inversion results comprises: Determining the horizontal position of the cavity based on the peak delay of the inter-station cross-correlation function according to the existence probability of the cavity; Converting the characteristic period into the burial depth of the cavity based on the abnormal phase velocity obtained by inversion of the dispersion curve; the horizontal position and the buried depth are spatially combined to obtain a three-dimensional positioning result of the cavity; constructing a comprehensive index according to the scattering effect intensity, the wave velocity amplitude reduction and the frequency dispersion abnormal amplitude; and based on the comprehensive index, demarcating and grading the range and risk of the cavity influence area, and outputting a structured detection report, wherein the structured detection report comprises the three-dimensional position, the influence range and the risk grade of the cavity.
8. The method of claim 7, wherein the cross-correlation function is formulated as follows: ; Wherein, the Indicating the time delay between station i and station j I and j respectively represent the position indexes of two stations; a ground-pulse signal value representing the time t of station i; A ground pulse signal value representing the time t+τ of station j; n represents the signal length; And Representing the standard deviation of the signals of stations i and j.
9. An intelligent detection system for roadbed hollows, the system comprising: the broadband seismograph array is used for collecting the ground pulse signals of the target roadbed area; A signal processing module for performing the method of any of claims 1-8 to generate a hole detection result; And the visual terminal is used for visually displaying the cavity detection result, and the cavity detection result comprises spatial position or high-risk early warning information.
10. An electronic device, comprising: at least one processor; at least one memory for storing at least one program; the at least one program, when executed by the at least one processor, causes the at least one processor to implement the method of any one of claims 1 to 8.

Description

Roadbed cavity intelligent detection method and relevant equipment based on multi-feature fusion and deep learning Technical Field The application relates to the technical field of roadbed cavity detection, in particular to a roadbed cavity intelligent detection method based on multi-feature fusion and deep learning and related equipment. Background In the related art, along with the acceleration of the urban process, the construction scale of infrastructures such as roads, railways and the like is continuously enlarged, and the stability and the safety of roadbed structures become important factors affecting public traffic safety. The existence of the cavity below the roadbed can cause the pavement to subside, crack and even collapse, and seriously threaten the driving safety and the engineering life. Therefore, early high-precision detection of voids is an important research direction in the field of infrastructure maintenance. The traditional detection method has certain limitation, the existing pulse detection depends on manual interpretation of spectrum characteristics (such as H/V spectrum ratio), is easily influenced by subjective judgment, and has high omission rate on small-scale cavities. The prior art uses single feature analysis, and only uses frequency domain features to hardly distinguish cavity signals from interference sources such as underground pipelines, rock stratum interfaces and the like. Urban roadbed vibration noise can mask weak cavity signals, and the traditional filtering method causes effective signal loss. The prior art has weak feature expression capability, and a single H/V spectrum bit sign cannot comprehensively represent signal nonlinearity and non-stationary characteristics caused by cavities, so that detection omission or false alarm is caused. The traditional method relies on manual feature extraction and threshold judgment, so that the automatic requirement of large-scale road inspection is difficult to meet, and the real-time processing capability is insufficient. In addition, the method based on fixed parameters or rules is difficult to adapt to the hole detection requirements under different geological conditions, and the model generalization is poor. In summary, the technical problems in the related art are to be improved. Disclosure of Invention The embodiment of the application mainly aims to provide an intelligent roadbed cavity detection method and related equipment based on multi-feature fusion and deep learning, aiming at improving the accuracy, efficiency and intelligent degree of roadbed cavity detection. In order to achieve the above object, an aspect of the embodiments of the present application provides an intelligent detection method for a roadbed cavity based on multi-feature fusion and deep learning, the method comprising: obtaining ground pulse signal data of a target roadbed area; performing frequency domain filtering and self-adaptive noise suppression on the ground pulse signal data to form a preprocessing signal sequence; Based on a physical effect model of the cavity, extracting characteristics of the preprocessing signal sequence to generate a multi-dimensional characteristic set; performing feature fusion and dimension reduction on the multi-dimensional feature set to obtain a fusion feature vector; optimizing the deep learning cavity recognition model based on the fusion feature vector; inputting the preprocessed signal sequence of the road section to be tested into the optimized deep learning cavity recognition model to obtain the probability of cavity existence; based on the probability of existence of the cavity, combining the cross-correlation analysis and the dispersion curve inversion result, completing space positioning and risk assessment of the cavity, and obtaining a cavity detection result, wherein the cavity detection result comprises space position or high risk early warning information. In some embodiments, the physical effect model comprises: A scattering effect sub-model for performing Gaussian attenuation modeling on scattering intensity in the pre-processing signal sequence based on the distance change between the station and the cavity center; a resonance effector model for identifying energy aggregation characteristics of specific frequency components in the pre-processed signal sequence based on physical property differences between the cavity and surrounding medium; And the wave velocity abnormal effector model characterizes low-speed deviation of the phase velocity in the preprocessing signal sequence in a target frequency band based on equivalent stiffness change caused by the cavity. In some embodiments, the scattering effector model is formulated as follows: ; Wherein, the Is shown in the firstThe intensity of scattering effects observed at the individual stations; Representing the maximum effect amplitude at the center of the cavity; Represent the first The location of the individual stations; representing the posi