Nondestructive detection method and system for pavement hidden diseases

Abstract

The invention discloses a pavement hidden disease nondestructive testing method and system, and the method comprises the steps: carrying out the alignment of electromagnetic wave, sound wave and thermal field signals through a space-time synchronization algorithm for a first signal set, and integrating multi-modal data through a weighted fusion model, so as to obtain a second fusion data set; for the first feature set, if the feature value is lower than a preset threshold value, the environmental noise interference is suppressed through an adaptive filter, and a second feature set is obtained; deep disease features are extracted from the second feature set, and a generative adversarial network is adopted to perform high-resolution reconstruction on small holes and crack extension to obtain a third feature set; and for the third feature set, generating a road deep structure image by adopting a three-dimensional imaging algorithm, and determining positions and forms of small holes and crack propagation through an image segmentation technology to obtain a first disease distribution diagram. According to the invention, the accuracy and efficiency of road deep disease detection can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of nondestructive detection of hidden pavement defects, and in particular relates to a nondestructive detection method and system for hidden pavement defects based on multi-source information fusion. Background Art

[0002] Nondestructive testing of road infrastructure is a key area for ensuring traffic safety and extending the service life of roads. With the continuous increase in traffic loads and the aging of roads, hidden pavement defects such as pavement voids, interlayer voids, and crack expansion are increasingly becoming a major threat to road collapse and structural damage. Efficient and accurate detection technologies are crucial for the timely detection and repair of these defects. However, existing commonly used drilling sampling methods have problems such as high destructiveness, low efficiency, and inability to identify deep-seated defects; ground-penetrating radar is susceptible to electromagnetic interference from underground pipelines and aquifers, and data interpretation is highly dependent on professional experience; ultrasonic testing requires coupling agents, is slow, and has difficulty in achieving large-scale coverage; visual and infrared inspections can only identify surface defects and cannot detect deep structural anomalies; traditional data processing algorithms are insufficient in the fusion of multi-source heterogeneous data, resulting in a high rate of misjudgment and significant limitations in practical applications.

[0003] These limitations reveal the core challenges facing this field. The effective collection and fusion of multimodal data is the primary difficulty. Due to the large differences in the signal characteristics of different physical fields, it is difficult to achieve synchronous acquisition and complementary analysis. The lack of data fusion further leads to inaccurate feature extraction of defects, especially in complex geological environments, where environmental noise and interference can significantly reduce detection resolution. The inaccuracy of feature extraction directly affects the ability to identify deep-seated defects, especially for the detection of small cavities or early cracks. Existing technologies often find it difficult to achieve sufficient depth and accuracy. The lack of mobility and automation of the detection platform also limits the large-scale application of the technology. Manual operation or semi-automated equipment can hardly meet the needs of long-distance and continuous detection.

[0004] Therefore, how to develop a detection technology that can efficiently integrate multimodal physical field data, accurately extract disease characteristics, achieve deep and high-resolution detection, and realize large-scale application through an automated platform has become a key issue in the field of road non-destructive testing. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a non-destructive detection method and system for hidden road defects, which can effectively improve the accuracy and efficiency of deep road defect detection and provide important technical support for road maintenance and safety management.

[0006] On the one hand, to achieve the above-mentioned object, the present invention provides a non-destructive detection method for hidden road surface defects, comprising: acquiring electromagnetic wave signals, acoustic wave signals, and thermal field signals to form multi-modal physical field signal data;

[0007] Preprocessing the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively to generate a first signal set;

[0008] performing spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal aligned signal set, and performing weighted fusion on the spatiotemporal aligned signal set to generate a second fused data set;

[0009] Extracting disease features from the second fused data set to generate a first feature set;

[0010] performing noise suppression on the first feature set to generate a second feature set;

[0011] extracting deep disease features from the second feature set, reconstructing the deep disease features at high resolution, and generating a third feature set;

[0012] Performing three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map;

[0013] Based on the first disease distribution map, an automated detection path is planned, and the multimodal physical field signal data collected in real time is compared and analyzed with historical data to determine the changing trend of disease distribution and generate a final disease analysis report.

[0014] Optionally, preprocessing the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal separately includes:

[0015] Performing Fourier transform on the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal to generate a frequency domain signal set;

[0016] Using a wavelet transform denoising algorithm to remove high-frequency noise from the frequency domain signal set to generate a denoised signal set;

[0017] The denoised signal set is normalized using a z-score normalization method to generate a normalized signal set.

[0018] Optionally, performing spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal alignment signal set includes:

[0019] Based on the highest sampling frequency, the first signal set is linearly interpolated to fill in the missing data of the time series to generate a time synchronization signal set;

[0020] The time synchronization signal set is corrected for the deviation of the sensor spatial sampling points by nearest neighbor interpolation to generate a time-space alignment signal set.

[0021] Optionally, performing weighted fusion on the spatiotemporally aligned signal set to generate a second fused data set includes:

[0022] Calculating the modal weights of the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal by a principal component analysis algorithm;

[0023] Performing weighted summation on the spatiotemporal alignment signal set according to the modal weights to generate a weighted fusion signal set;

[0024] Perform mean normalization and sliding window smoothing processing on the weighted fusion signal set to generate a second fusion data set.

[0025] Optionally, extracting disease features from the second fused data set to generate a first feature set includes:

[0026] Performing multi-scale feature decomposition on the second fused data set using a convolutional neural network;

[0027] The feature weights corresponding to small holes and crack expansions are enhanced through the attention mechanism to generate the first feature set.

[0028] Optionally, performing noise suppression on the first feature set to generate a second feature set includes:

[0029] When the eigenvalue of the first feature set is lower than a preset threshold, a Kalman filter is used to adaptively suppress the high-frequency noise component;

[0030] The noise-suppressed feature set is decomposed into multiple scales through wavelet transform, the energy distribution of low-frequency components and high-frequency components is extracted, and the second feature set is generated.

[0031] Optionally, extracting deep-seated disease features from the second feature set and reconstructing the deep-seated disease features at high resolution to generate a third feature set includes:

[0032] Performing wavelet transform decomposition on the second feature set to extract low-frequency components and high-frequency components to generate a preliminary disease feature set;

[0033] The contribution rate of each component is calculated through principal component analysis, and the components with high contribution rate are weighted and fused to generate the optimized disease feature set;

[0034] The optimized disease feature set is reconstructed at high resolution using a generative adversarial network to generate a third feature set.

[0035] Optionally, performing three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map includes:

[0036] reconstructing the third feature set into a preliminary three-dimensional image of the road deep structure;

[0037] Extracting the boundary of the diseased area in the three-dimensional image to generate a first segmented image;

[0038] Based on the pixel intensity of the first segmented image, the center coordinates and boundary points of the diseased area are determined by a spatial clustering algorithm to generate a first disease distribution map.

[0039] On the other hand, to achieve the above-mentioned purpose, the present invention provides a non-destructive detection system for hidden road defects, comprising:

[0040] Signal acquisition module, signal preprocessing module, spatiotemporal synchronization and fusion module, feature extraction and enhancement module, noise suppression module, high-resolution reconstruction module, three-dimensional imaging and segmentation module, and path planning and dynamic analysis module;

[0041] The signal acquisition module is used to acquire electromagnetic wave signals, acoustic wave signals, and thermal field signals to form multimodal physical field signal data;

[0042] The signal preprocessing module is used to preprocess the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively to generate a first signal set;

[0043] The spatiotemporal synchronization and fusion module is configured to perform spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal aligned signal set, and perform weighted fusion on the spatiotemporal aligned signal set to generate a second fused data set;

[0044] The feature extraction and enhancement module is used to extract disease features based on the second fused data set to generate a first feature set;

[0045] The noise suppression module is configured to perform noise suppression on the first feature set to generate a second feature set;

[0046] The high-resolution reconstruction module is used to extract deep-seated disease features based on the second feature set, perform high-resolution reconstruction on the deep-seated disease features, and generate a third feature set;

[0047] The three-dimensional imaging and segmentation module is used to perform three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map;

[0048] The path planning and dynamic analysis module is used to plan an automated detection path based on the first disease distribution map, compare and analyze the multimodal physical field signal data collected in real time with historical data, determine the trend of disease distribution changes, and generate a final disease analysis report.

[0049] Technical Effects of the Invention: The present invention discloses a nondestructive detection method and system for hidden pavement defects. Through multimodal physical field signal acquisition and fusion technology, high-precision detection of small cavities and crack expansion within the pavement is achieved. First, multi-source data such as ground-penetrating radar, ultrasonic waves, and infrared thermal imaging are preprocessed and fused. Then, deep learning algorithms are used to extract defect features, and high-resolution reconstruction is performed using a generative adversarial network. Finally, three-dimensional imaging and image segmentation technology are used to determine the location and morphology of the defect, and continuous detection and real-time analysis are achieved in conjunction with an automated detection platform. The present invention can effectively improve the accuracy and efficiency of deep pavement defect detection, providing important technical support for road maintenance and safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0051] Figure 1 This is a flow chart of a nondestructive detection method for hidden road surface defects according to an embodiment of the present invention;

[0052] Figure 2 The figure is a schematic structural diagram of a nondestructive detection system for hidden road surface defects according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0054] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0055] like Figure 1 As shown, this embodiment provides a non-destructive detection method for hidden road defects, including:

[0056] Acquire electromagnetic wave signals, acoustic wave signals, and thermal field signals to form multimodal physical field signal data;

[0057] Preprocessing the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively to generate a first signal set;

[0058] performing spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal aligned signal set, and performing weighted fusion on the spatiotemporal aligned signal set to generate a second fused data set;

[0059] Extracting disease features from the second fused data set to generate a first feature set;

[0060] performing noise suppression on the first feature set to generate a second feature set;

[0061] extracting deep disease features from the second feature set, reconstructing the deep disease features at high resolution, and generating a third feature set;

[0062] Performing three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map;

[0063] Based on the first disease distribution map, an automated detection path is planned, and the multimodal physical field signal data collected in real time is compared and analyzed with historical data to determine the changing trend of disease distribution and generate a final disease analysis report.

[0064] Furthermore, preprocessing the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively includes:

[0065] Performing Fourier transform on the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal to generate a frequency domain signal set;

[0066] Using a wavelet transform denoising algorithm to remove high-frequency noise from the frequency domain signal set to generate a denoised signal set;

[0067] The denoised signal set is normalized using a z-score normalization method to generate a normalized signal set.

[0068] Specifically, the Fourier transform converts the initial signal set into a frequency domain signal set. The Fourier transform extracts frequency components from electromagnetic signals, reflecting the characteristic frequencies of the pipe material and depth. For example, high frequencies may correspond to metal pipes, while low frequencies may correspond to plastic pipes. Frequency domain analysis of acoustic signals can identify specific frequency peaks caused by cracks, such as 500 Hz, which may correspond to small cracks. Frequency domain conversion of thermal signals can highlight the periodic characteristics of temperature variations. For example, a leak may manifest as low-frequency temperature fluctuations. The frequency domain signal set provides clear frequency characteristics for subsequent denoising. The wavelet transform denoising algorithm processes the frequency domain signal set. In electromagnetic signals, environmental noise may manifest as high-frequency random fluctuations. The wavelet transform decomposes the signal, retaining low-frequency pipeline characteristics and removing high-frequency noise. In acoustic signals, background echoes may interfere with crack detection. The wavelet transform extracts mid-frequency crack characteristics and suppresses low-frequency background noise. In thermal signals, ambient temperature fluctuations may mask leak characteristics. The wavelet transform highlights localized high temperature characteristics, resulting in a more accurate denoised signal. The denoised signal set significantly improves the signal-to-noise ratio and enhances signal reliability.

[0069] Further, performing spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal alignment signal set includes:

[0070] Based on the highest sampling frequency, the first signal set is linearly interpolated to fill in the missing data of the time series to generate a time synchronization signal set;

[0071] The time synchronization signal set is corrected for the deviation of the sensor spatial sampling points by nearest neighbor interpolation to generate a time-space alignment signal set.

[0072] Specifically, for underground pipeline inspection, the first signal set includes electromagnetic wave signals acquired by a ground-penetrating radar (GPR), acoustic wave signals collected by an ultrasonic sensor, and thermal field signals recorded by an infrared thermal imaging device. A timestamp alignment algorithm ensures the synchronization of these signals along the time axis. The GPR collects 100 electromagnetic wave intensity data points per second, the ultrasonic sensor records 80 echo time data points per second, and the thermal field signal generates 120 temperature values ​​per second. Due to different sampling frequencies of the devices, timestamps may vary. The timestamp alignment algorithm uses the highest sampling frequency as the basis, for example, every 10 millisecond as a time node, to uniformly align all signals. Linear interpolation fills in missing data. For example, if acoustic wave data is missing at a time node, a linear estimate can be made based on the echo times of the preceding and following time points to generate a complete set of time-synchronized signals. A spatial coordinate mapping algorithm aligns the spatial dimensions of the signals. The GPR signal corresponds to the location of the pipeline at a depth of 1 meter underground, the ultrasonic signal reflects the spatial distribution of internal defects within the pipeline, and the thermal field signal records the spatial variation of the pipeline surface temperature. If the sampling point position deviates, for example, the sampling point of the ground penetrating radar deviates by 5 cm from the actual pipeline center, the nearest neighbor interpolation method can assign the values ​​of the neighboring sampling points to the correct position and generate a spatiotemporally aligned signal set.

[0073] Furthermore, performing weighted fusion on the spatiotemporal aligned signal set to generate a second fused data set includes:

[0074] Calculating the modal weights of the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal by a principal component analysis algorithm;

[0075] Performing weighted summation on the spatiotemporal alignment signal set according to the modal weights to generate a weighted fusion signal set;

[0076] Perform mean normalization and sliding window smoothing processing on the weighted fusion signal set to generate a second fusion data set.

[0077] Furthermore, extracting disease features from the second fused data set to generate a first feature set includes:

[0078] Performing multi-scale feature decomposition on the second fused data set using a convolutional neural network;

[0079] The feature weights corresponding to small holes and crack expansions are enhanced through the attention mechanism to generate the first feature set.

[0080] Specifically, in an underground pipeline inspection scenario, time series data is extracted from the second fused dataset, containing the combined intensity of electromagnetic, acoustic, and thermal signals. The principle of wavelet transform decomposition of time series is to decompose the signal into components of different frequencies, capturing local variations caused by pipeline defects. The Haar wavelet basis function is used to decompose the time series into high-frequency and low-frequency components. High-frequency components reflect rapidly changing defects such as cracks, while low-frequency components represent slowly changing overall pipeline conditions. In a particular pipeline inspection, the time series contains 100 fused intensity data points per second. The wavelet transform generates high-frequency components, which reveal sharp fluctuations caused by cracks, while low-frequency components reflect the stable trend of pipeline depth. Frequency domain energy distribution calculates the energy ratio of high- and low-frequency components to form a frequency domain feature set. For a certain section of pipeline, the high-frequency component energy is concentrated in the 10-20 Hz range, indicating crack vibration characteristics; the low-frequency component energy is concentrated in the 0-5 Hz range, reflecting the uniformity of the pipeline material. The frequency domain feature set records this energy distribution for subsequent analysis. A mean filter smoothes the high-frequency component to reduce noise interference. The average value is calculated using a window of five data points to generate a smoothed signal. A high-frequency component sequence is [8, 12, 10, 15, 7], which is reduced to [10, 11, 10.4] after mean filtering, reducing sharp noise. If the peak value of the smoothed signal exceeds a threshold of 10, subsequent analysis is triggered, indicating the presence of a significant defect. Fourier transform is used to extract spatial frequency components and generate a spatial frequency feature set. A peak value of 12 in a smoothed signal is detected by Fourier transform, which identifies a spatial frequency of 0.5 cycles / meter, corresponding to periodic cracks on the pipeline surface. The spatial frequency feature set records these components and reflects the spatial distribution of the defect. Principal component analysis is used to calculate the contribution of spatial frequency components and integrate components with high contributions. In one inspection, the 0.5 cycles / meter component contributed 60% and was weighted 0.6; the 0.2 cycles / meter component contributed 30% and was weighted 0.3. A weighted summation is used to generate an enhanced feature set, highlighting the main defect characteristics. A clustering algorithm performs pattern classification on the enhanced feature set to identify the defect type.

[0081] Furthermore, performing noise suppression on the first feature set to generate a second feature set includes:

[0082] When the eigenvalue of the first feature set is lower than a preset threshold, a Kalman filter is used to adaptively suppress the high-frequency noise component;

[0083] The noise-suppressed feature set is decomposed into multiple scales through wavelet transform, the energy distribution of low-frequency components and high-frequency components is extracted, and the second feature set is generated.

[0084] Specifically, in underground pipeline inspection scenarios, when the eigenvalues ​​of the first feature set fall below a preset threshold, the Kalman filter method can be used to adaptively suppress high-frequency noise components. The principle of Kalman filtering is to dynamically adjust the noise interference in the feature set through prediction and update steps, combining observed data and system models. In one inspection, the high-frequency component energy was concentrated in the 15-25Hz range, indicating crack vibration; the low-frequency component energy was in the 0-3Hz range, reflecting the stability of the pipeline depth. The energy feature set records these distributions and provides a basis for subsequent processing. Principal component analysis (PCA) calculates the contribution of each component in the energy feature set and fuses the high-contributing components to generate an enhanced feature set. PCA reduces dimensionality and extracts the components that contribute most to the variation in the feature set. For an energy feature set containing 10 components, PCA shows that the 15Hz frequency component contributes 50%, and the 5Hz frequency component contributes 30%. Weights of 0.5 and 0.3 were assigned to these components, respectively, and weighted fusion was used to generate an enhanced feature set that highlights crack-related features. This weighting method effectively reduces the interference of minor components. If the intra-class variance of the enhanced feature set exceeds the preset threshold of 0.2, the K-means clustering algorithm is used for pattern classification to obtain the second feature set. The principle of K-means clustering is to assign feature points to the cluster with the smallest intra-class variance to identify defect patterns.

[0085] Furthermore, deep-seated disease features are extracted from the second feature set, and the deep-seated disease features are reconstructed at high resolution to generate a third feature set including:

[0086] Performing wavelet transform decomposition on the second feature set to extract low-frequency components and high-frequency components to generate a preliminary disease feature set;

[0087] The contribution rate of each component is calculated through principal component analysis, and the components with high contribution rate are weighted and fused to generate the optimized disease feature set;

[0088] The optimized disease feature set is reconstructed at high resolution using a generative adversarial network to generate a third feature set.

[0089] Furthermore, performing three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map includes:

[0090] reconstructing the third feature set into a preliminary three-dimensional image of the road deep structure;

[0091] Extracting the boundary of the diseased area in the three-dimensional image to generate a first segmented image;

[0092] Based on the pixel intensity of the first segmented image, the center coordinates and boundary points of the diseased area are determined by a spatial clustering algorithm to generate a first disease distribution map.

[0093] Specifically, in the field of underground road inspection, processing starting from the third feature set involves stereomicroscopy algorithms, level set segmentation algorithms, spatial clustering algorithms, and geometric fitting algorithms, focusing on the 3D reconstruction and feature extraction of deep road defects. The following technical topic is analyzed from multiple perspectives through a theoretical introduction, implementation methods, and examples, ensuring a rigorous logic and a focus on road inspection scenarios. The stereomicroscopy algorithm reconstructs the 2D feature data from the third feature set into a 3D image by simulating the imaging principle of a stereo microscope, generating a preliminary 3D image of the deep road structure. In principle, this algorithm utilizes multi-angle feature projections combined with depth information to construct a 3D point cloud. In one embodiment, a first 3D image contains voids and cracks. The level set algorithm, driven by pixel intensity gradients, segments the void boundaries, generating a first segmented image in which the void areas have clear boundaries and the crack extension areas exhibit elongated textures. This segmentation method highlights the morphological characteristics of the defect area, providing a precise foundation for subsequent location determination. The spatial clustering algorithm calculates the center coordinates and boundary points of the defect area based on the pixel intensities in the first segmented image to determine the spatial location of the defect. In principle, the algorithm identifies the geometric center and boundaries of the diseased area by grouping high-intensity pixels.

[0094] like Figure 2 As shown, this embodiment also provides a non-destructive detection system for hidden road defects, including:

[0095] Signal acquisition module, signal preprocessing module, spatiotemporal synchronization and fusion module, feature extraction and enhancement module, noise suppression module, high-resolution reconstruction module, three-dimensional imaging and segmentation module, and path planning and dynamic analysis module;

[0096] The signal acquisition module is used to acquire electromagnetic wave signals, acoustic wave signals, and thermal field signals to form multimodal physical field signal data;

[0097] The signal preprocessing module is used to preprocess the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively to generate a first signal set;

[0098] The spatiotemporal synchronization and fusion module is configured to perform spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal aligned signal set, and perform weighted fusion on the spatiotemporal aligned signal set to generate a second fused data set;

[0099] The feature extraction and enhancement module is used to extract disease features based on the second fused data set to generate a first feature set;

[0100] The noise suppression module is configured to perform noise suppression on the first feature set to generate a second feature set;

[0101] The high-resolution reconstruction module is used to extract deep-seated disease features based on the second feature set, perform high-resolution reconstruction on the deep-seated disease features, and generate a third feature set;

[0102] The three-dimensional imaging and segmentation module is used to perform three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map;

[0103] The path planning and dynamic analysis module is used to plan an automated detection path based on the first disease distribution map, compare and analyze the multimodal physical field signal data collected in real time with historical data, determine the trend of disease distribution changes, and generate a final disease analysis report.

[0104] This invention discloses a nondestructive detection method and system for hidden pavement defects. Through multimodal physical field signal acquisition and fusion technology, it achieves high-precision detection of small cavities and crack expansion within the pavement. First, multi-source data, including ground-penetrating radar, ultrasonic waves, and infrared thermal imaging, is preprocessed and fused. Then, deep learning algorithms are used to extract defect features, and high-resolution reconstruction is performed using a generative adversarial network. Finally, three-dimensional imaging and image segmentation techniques are used to determine the defect's location and morphology. Combined with an automated detection platform, continuous detection and real-time analysis are achieved. This invention can effectively improve the accuracy and efficiency of deep-seated pavement defect detection, providing important technical support for road maintenance and safety management.

[0105] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A non-destructive detection method for hidden road defects, characterized in that: include: Acquire electromagnetic wave signals, acoustic wave signals, and thermal field signals to form multimodal physical field signal data; Preprocessing the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively to generate a first signal set; performing spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal aligned signal set, and performing weighted fusion on the spatiotemporal aligned signal set to generate a second fused data set; Extracting disease features from the second fused data set to generate a first feature set; performing noise suppression on the first feature set to generate a second feature set; extracting deep disease features from the second feature set, reconstructing the deep disease features at high resolution, and generating a third feature set; Performing three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map; Based on the first disease distribution map, an automated detection path is planned, and the multimodal physical field signal data collected in real time is compared and analyzed with historical data to determine the changing trend of disease distribution and generate a final disease analysis report.

2. The nondestructive detection method for hidden road defects according to claim 1, characterized in that: Preprocessing the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively includes: Performing Fourier transform on the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal to generate a frequency domain signal set; Using a wavelet transform denoising algorithm to remove high-frequency noise from the frequency domain signal set to generate a denoised signal set; The denoised signal set is normalized using a z-score normalization method to generate a normalized signal set.

3. The nondestructive detection method for hidden road surface defects according to claim 1, characterized in that: Performing spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal alignment signal set includes: Based on the highest sampling frequency, the first signal set is linearly interpolated to fill in the missing data of the time series to generate a time synchronization signal set; The time synchronization signal set is corrected for the deviation of the sensor spatial sampling points by nearest neighbor interpolation to generate a time-space alignment signal set.

4. The nondestructive detection method for hidden road surface defects according to claim 1, characterized in that: Performing weighted fusion on the spatiotemporally aligned signal set to generate a second fused data set includes: Calculating the modal weights of the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal by a principal component analysis algorithm; Performing weighted summation on the spatiotemporal alignment signal set according to the modal weights to generate a weighted fusion signal set; Perform mean normalization and sliding window smoothing processing on the weighted fusion signal set to generate a second fusion data set.

5. The nondestructive detection method for hidden road surface defects according to claim 1, characterized in that: Extracting disease features from the second fused data set to generate a first feature set includes: Performing multi-scale feature decomposition on the second fused data set using a convolutional neural network; The feature weights corresponding to small holes and crack expansions are enhanced through the attention mechanism to generate the first feature set.

6. The nondestructive detection method for hidden road surface defects according to claim 1, characterized in that: Performing noise suppression on the first feature set to generate a second feature set includes: When the eigenvalue of the first feature set is lower than a preset threshold, a Kalman filter is used to adaptively suppress the high-frequency noise component; The noise-suppressed feature set is decomposed into multiple scales through wavelet transform, the energy distribution of low-frequency components and high-frequency components is extracted, and the second feature set is generated.

7. The nondestructive detection method for hidden road defects according to claim 1, characterized in that: Extracting deep-seated disease features from the second feature set and reconstructing the deep-seated disease features at high resolution to generate a third feature set includes: Performing wavelet transform decomposition on the second feature set to extract low-frequency components and high-frequency components to generate a preliminary disease feature set; The contribution rate of each component is calculated through principal component analysis, and the components with high contribution rate are weighted and fused to generate the optimized disease feature set; The optimized disease feature set is reconstructed at high resolution using a generative adversarial network to generate a third feature set.

8. The nondestructive detection method for hidden road surface defects according to claim 1, characterized in that: Performing three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map includes: reconstructing the third feature set into a preliminary three-dimensional image of the road deep structure; Extracting the boundary of the diseased area in the three-dimensional image to generate a first segmented image; Based on the pixel intensity of the first segmented image, the center coordinates and boundary points of the diseased area are determined by a spatial clustering algorithm to generate a first disease distribution map.

9. A system for nondestructive detection of hidden road surface defects according to any one of claims 1 to 8, characterized in that: include: Signal acquisition module, signal preprocessing module, spatiotemporal synchronization and fusion module, feature extraction and enhancement module, noise suppression module, high-resolution reconstruction module, three-dimensional imaging and segmentation module, and path planning and dynamic analysis module; The signal acquisition module is used to acquire electromagnetic wave signals, acoustic wave signals, and thermal field signals to form multimodal physical field signal data; The signal preprocessing module is used to preprocess the electromagnetic wave signal, the acoustic wave signal, and the thermal field signal respectively to generate a first signal set; The spatiotemporal synchronization and fusion module is configured to perform spatiotemporal synchronization processing on the first signal set to generate a spatiotemporal aligned signal set, and perform weighted fusion on the spatiotemporal aligned signal set to generate a second fused data set; The feature extraction and enhancement module is used to extract disease features based on the second fused data set to generate a first feature set; The noise suppression module is configured to perform noise suppression on the first feature set to generate a second feature set; The high-resolution reconstruction module is used to extract deep-seated disease features based on the second feature set, perform high-resolution reconstruction on the deep-seated disease features, and generate a third feature set; The three-dimensional imaging and segmentation module is used to perform three-dimensional imaging and segmentation processing on the third feature set to generate a first disease distribution map; The path planning and dynamic analysis module is used to plan an automated detection path based on the first disease distribution map, compare and analyze the multimodal physical field signal data collected in real time with historical data, determine the trend of disease distribution changes, and generate a final disease analysis report.

Images