Method and device for rapidly detecting effective hard layer of high fill roadbed

Abstract

The invention provides a rapid detection method and device for an effective hard layer of a high fill roadbed, and relates to the technical field of roadbed rolling compaction. Performing time and space alignment on the roadbed multi-modal data to obtain aligned multi-modal data; performing feature extraction on the aligned multi-modal data, and performing unsupervised dimension reduction processing on the extracted features to obtain a multi-parameter fusion vector; the multi-parameter fusion vector is analyzed in real time according to edge nodes, and the compaction quality is rapidly evaluated in combination with a lightweight machine learning model to obtain a compaction degree score; and training the roadbed multi-modal data according to a lightweight hybrid neural network, predicting settlement, and carrying out roadbed detection according to the roadbed settlement trend and the compaction degree score. The method solves the problem that roadbed improvement measures lack pertinence and effectiveness and cannot fundamentally solve the problem of roadbed quality.

Description

Technical Field

[0001] This invention relates to the field of roadbed compaction technology, and more specifically, to a method and apparatus for rapid detection of the effective hard layer in high embankment roadbeds. Background Technology

[0002] Within the current roadbed compaction technology system, single indicators such as vibration value, density, and vibration compaction degree are typically used to evaluate the condition of the roadbed hard layer. However, this evaluation method based on a single indicator has significant and undeniable limitations. From a mechanical perspective, the roadbed hard layer, as a crucial load-bearing layer of the road structure, encompasses multiple dimensions of mechanical properties, including elastic modulus, shear strength, and Poisson's ratio. A single indicator can only partially reflect some of these characteristics and cannot fully depict the mechanical response characteristics of the roadbed hard layer under complex stress states. In practical engineering applications, this easily leads to a "treating the symptoms but not the root cause" dilemma, resulting in roadbed remediation measures lacking specificity and effectiveness, and failing to fundamentally solve the problem of roadbed quality.

[0003] There is an urgent need for a rapid detection method and device for the effective hard layer of high embankment subgrade, which solves the problem that subgrade treatment measures lack specificity and effectiveness and cannot fundamentally solve the subgrade quality problem. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for rapid detection of the effective hard layer in high-fill roadbeds, thereby improving the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0005] Firstly, this application provides a rapid detection method for the effective hard layer of a high-fill roadbed, including:

[0006] Acquire multimodal data of the roadbed, which includes multi-directional vibration signals, environmental features, multispectral images, and pressure signals;

[0007] The roadbed multimodal data is time-space aligned to obtain aligned multimodal data;

[0008] Feature extraction is performed on the aligned multimodal data, and unsupervised dimensionality reduction is applied to the extracted features to obtain a multi-parameter fusion vector;

[0009] Based on real-time analysis of the multi-parameter fusion vector at the edge nodes, and combined with a lightweight machine learning model, the compaction quality is quickly evaluated to obtain a compaction score.

[0010] The subgrade multimodal data is trained using a lightweight hybrid neural network to predict settlement, and the subgrade is detected by subgrade settlement trend and compaction score.

[0011] Secondly, this application also provides a rapid detection device for the effective hard layer of a high embankment subgrade, comprising:

[0012] The acquisition module is used to acquire roadbed multimodal data, which includes multi-directional vibration signals, environmental features, multispectral images, and pressure signals.

[0013] An alignment module is used to perform temporal and spatial alignment on the roadbed multimodal data to obtain aligned multimodal data;

[0014] The extraction module is used to extract features from the aligned multimodal data and obtain a multi-parameter fusion vector by performing unsupervised dimensionality reduction on the extracted features.

[0015] The evaluation module is used to analyze the multi-parameter fusion vector in real time based on the edge nodes, and combine it with a lightweight machine learning model to quickly evaluate the compaction quality and obtain a compaction score.

[0016] The prediction module is used to train the subgrade multimodal data based on a lightweight hybrid neural network and predict settlement, and to detect the subgrade through subgrade settlement trend and compaction score.

[0017] The beneficial effects of this invention are as follows:

[0018] This invention utilizes multimodal data of the roadbed, comprehensively considering various physical quantities and environmental factors, to conduct a comprehensive and accurate assessment of roadbed compaction quality. Compared with traditional detection methods, this invention effectively overcomes the limitations of traditional methods in terms of detection accuracy and real-time performance. Unsupervised dimensionality reduction comprehensively reflects the physical characteristics of the roadbed, effectively extracting key features and reducing data redundancy. Furthermore, by combining a lightweight machine learning model with a hybrid neural network, it achieves rapid and accurate real-time assessment and prediction of roadbed conditions. The lightweight machine learning model can efficiently process large-scale data, while the hybrid neural network further improves the accuracy and efficiency of assessment and prediction. In summary, this invention solves the problem of roadbed remediation measures lacking specificity and effectiveness, failing to fundamentally address the issue of roadbed quality.

[0019] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of a rapid detection method for the effective hard layer of a high embankment subgrade as described in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of a rapid detection device for the effective hard layer of a high embankment roadbed, as described in an embodiment of the present invention.

[0023] The markings in the figure are: 800, a rapid detection device for the effective hard layer of a high embankment subgrade; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0025] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0026] Example 1:

[0027] This embodiment provides a rapid detection method for the effective hard layer of a high-fill roadbed.

[0028] See Figure 1 The figure shows that the method includes steps S1 to S5, including:

[0029] S1: Acquire roadbed multimodal data, which includes multi-directional vibration signals, environmental features, multispectral images, and pressure signals;

[0030] In this step, the roadbed multimodal data is acquired through sensor measurements, taking into account a variety of physical quantities and environmental factors.

[0031] Preferably, the multi-directional vibration signal is obtained by measuring a high-precision triaxial accelerometer, the pressure signal is obtained by measuring a distributed pressure sensor, the environmental characteristics are obtained by measuring an environmental sensor, and the multispectral image is obtained by measuring visible light (RGB) and near-infrared (NIR, wavelength 800-900nm).

[0032] S2: Perform temporal and spatial alignment on the roadbed multimodal data to obtain aligned multimodal data;

[0033] To clarify the specific method for acquiring aligned multimodal data, step S2 includes S21 to S24, specifically:

[0034] S21: Time-stamp align the pressure signal and the multispectral image to obtain time-aligned data;

[0035] In this step, the distributed pressure sensor, the high-precision triaxial accelerometer, the environmental sensor, and the multispectral image are time-stamped to obtain time-aligned data, ensuring that the time error between the sensor and the UAV data acquisition is <1ms. This solves the problem of asynchronous sampling frequencies of multiple sensors and ensures the consistency of vibration, pressure signals, and the multispectral image in the time dimension.

[0036] S22: Assign coordinate values ​​to multiple pixels in the multispectral image and establish a two-dimensional coordinate system to obtain a spatially labeled image;

[0037] In this step, each pixel in the multispectral image is assigned a unique coordinate value, and a two-dimensional coordinate system is established to correspond the pixel with its actual spatial location, thereby obtaining a spatially labeled image.

[0038] S23: Based on the spatially labeled image, spatially aligned data is obtained by spatially aligning the coordinates of the measurement points of the pressure signal with the pixel coordinates of the multispectral image;

[0039] In this step, the spatial alignment is a millimeter-level spatial matching.

[0040] S24: Based on the time-aligned data and the spatial-aligned data, environmental compensation is performed and the data is fused to construct aligned multimodal data.

[0041] S3: Perform feature extraction on the aligned multimodal data, and obtain a multi-parameter fusion vector by performing unsupervised dimensionality reduction on the extracted features;

[0042] To clarify the specific method for obtaining the multi-parameter fusion vector, step S3 includes S31 to S35, specifically:

[0043] S31: Obtain real-time operating condition data;

[0044] S32: Extract features from the aligned multimodal data to obtain an initial feature set;

[0045] To clarify the specific method for obtaining the initial feature set, step S32 includes steps S321 to S325, which are as follows:

[0046] S321: Aggregate the pressure signal to generate grid-level pressure values ​​and perform feature calculations to obtain pressure features;

[0047] In this step, the pressure characteristics include time-domain statistical characteristics, spatial gradient characteristics, and dynamic response characteristics;

[0048] First, the pressure signal is aggregated to generate pressure grid cells, and the grid-level pressure value of each grid cell is calculated. The standard value and coefficient of variation of the stress are calculated using the grid-level pressure value to obtain the time-domain statistical characteristics.

[0049] Furthermore, spatial gradient features are extracted by calculating the pressure gradient between adjacent grid cells.

[0050] Subsequently, the pressure waveform is extracted from the spatial gradient features, and a Fourier transform is performed on the extracted pressure waveform to obtain its spectrum. The maximum value point is then located in the spectrum to determine the peak frequency. Furthermore, the envelope of the pressure waveform is subjected to exponential fitting, and the parameters are fitted using a nonlinear least squares method to obtain the decay rate.

[0051] Finally, the peak frequency and decay rate of the pressure waveform are extracted to characterize the dynamic response properties of the material.

[0052] S322: Perform wavelet packet decomposition and frequency band quantization on the multi-directional vibration signal to obtain vibration characteristics;

[0053] In this step, the vibration characteristics include joint time-frequency domain characteristics and frequency domain characteristics. First, wavelet packet decomposition is performed on the multi-directional vibration signal to obtain multiple frequency bands. Then, the energy proportion and entropy value of each frequency band are calculated to obtain the joint time-frequency domain characteristics; the quantized signal of the entropy value indicates the uniformity of energy distribution in different frequency bands. The larger the entropy value, the more dispersed the energy distribution, corresponding to a more complex compaction state.

[0054] The entropy value expression is:

[0055] H energy =-∑ i p i ·ln(p i (1)

[0056] In equation (1) above, H energy p represents the entropy value. i ln(p) represents the energy percentage of the i-th frequency band. i ) represents p i The natural logarithm;

[0057]

[0058] In equation (2) above, p i E represents the energy percentage of the i-th frequency band. i This represents the energy of the i-th frequency band.

[0059] In addition, the spectral characteristics of the multi-directional vibration signal are extracted based on the short-time Fourier transform, and the energy proportion of the 0-50Hz frequency band (E0-50Hz) is analyzed in particular to reflect the low-frequency vibration response capability of the roadbed, thus obtaining the frequency domain characteristics.

[0060] S323: Calculate the mean and standard deviation of the pressure characteristic and the vibration characteristic based on the environmental characteristics to obtain the environmental correction factor;

[0061] S324: Analyze the multispectral image according to the target detection algorithm, identify roadbed cracks and calculate crack density to obtain image features;

[0062] In this step, the pixel coordinates, length, and width of the cracks are identified in real time based on the YOLOv7 model, and the crack density is calculated. The cracks are accurately located in the image using precise pixel coordinates.

[0063] A depth map of the settlement area is generated based on binocular vision and lidar point cloud data, and the volume loss rate is calculated.

[0064] The volume loss rate is:

[0065]

[0066] In equation (3) above, V loss D represents the volume loss rate. settlement,i D represents the depth of the i-th settlement zone. ref Indicates the reference elevation, A i V represents the pixel area of ​​the i-th settlement region. total Indicates the total detection volume;

[0067] Preferably, the target detection algorithm is the YOLOv7 model;

[0068] S325: Integrate the environmental correction factor and the image features to obtain an initial feature set.

[0069] S33: The high-dimensional features in the initial feature set are reduced in dimensionality using an unsupervised machine learning model to obtain the dimensionality-reduced feature vector;

[0070] In this step, the high-dimensional features in the initial feature set are reduced in dimensionality using Principal Component Analysis (PCA) in the unsupervised machine learning model. The first three principal components are extracted, retaining more than 90% of the information, reducing computational complexity, and preserving the main information of the data.

[0071] S34: Based on the real-time operating data, the preset feature importance coefficients are dynamically adjusted to obtain dynamically weighted features;

[0072] In this step, the preset feature importance coefficients are adjusted based on Min-Max standardization and real-time operating data (temperature, humidity, etc.) to eliminate dimensional differences and ensure that different features are compared and calculated within the same framework.

[0073] Preferably, the preset feature importance coefficient can be set based on experience or historical data.

[0074] S35: Construct a multi-parameter fusion vector based on the dynamically weighted features and the dimensionality-reduced feature vector.

[0075] S4: Based on the real-time analysis of the multi-parameter fusion vector at the edge nodes, and combined with a lightweight machine learning model, the compaction quality is quickly evaluated to obtain a compaction score.

[0076] To clarify the specific method for obtaining the compaction score, step S4 includes S41 to S44, specifically:

[0077] S41: Calculate the weights of the multi-parameter fusion vector based on the edge nodes and the online learning algorithm, and generate a dynamic weight vector;

[0078] In this step, the edge nodes calculate the weights of the multi-parameter fusion vector in real time according to preset rules (number of compaction passes, material type, construction stage) to obtain a dynamic weight vector;

[0079] Among them, the preset rules are weight adjustment strategies obtained by training on historical data based on online learning algorithms;

[0080] S42: Based on a long short-term memory network, perform time-series modeling of the multi-directional vibration signal and extract the hidden state vector;

[0081] In this step, a hybrid algorithm is constructed by integrating Long Short-Term Memory (LSTM) networks and lightweight convolutional neural networks. This algorithm is used to perform temporal modeling of the multi-directional vibration signals and extract hidden state vectors. This method can effectively capture the temporal dependence characteristics of vibration signals while reducing model complexity and computational resource consumption.

[0082] The hidden state vector expression is:

[0083] h t =LSTM[X t ,h t-1 ]→h t =LSTM t+1 [X t ,h t-1 N t+1 (4)

[0084] In equation (4) above, h t X is the hidden state vector output by the LSTM at time step t; t h is the input feature vector at time step t. t-1 N represents the output hidden state vector of the LSTM at time step t-1. t+1 This indicates the number of neurons that are dynamically adjusted.

[0085] S43: Detect abnormal regions of the multi-directional vibration signal based on the dynamic weight vector, and perform spatiotemporal alignment of the multispectral image through the abnormal regions to extract spatiotemporal features;

[0086] To clarify the specific method for obtaining spatiotemporal features, step S43 includes steps S431 to S433, specifically:

[0087] S431: The detection threshold is obtained by integrating the dynamic weight vector with the preset real-time environmental compensation. By automatically adjusting the detection threshold for anomalies, a multi-parameter adaptive threshold matrix is ​​obtained.

[0088] In this step, the dynamic weight vector is integrated with the preset real-time environmental compensation; the threshold for anomaly detection is automatically adjusted according to the integration result to obtain a multi-parameter adaptive threshold matrix, which accurately reflects the normal and abnormal states in the current environment.

[0089] S432: Based on the multi-parameter adaptive threshold matrix, anomaly identification is performed on the multi-directional vibration signal to obtain the abnormal region;

[0090] In this step, the multi-directional vibration signal is anomaly identified based on the multi-parameter adaptive threshold matrix. By comparing the multi-directional vibration signal with the multi-parameter adaptive threshold matrix, abnormal regions are determined, and abnormal regions in the vibration signal are quickly identified.

[0091] S433: Based on the spatiotemporal alignment of the abnormal region with the crack and subsidence regions in the multispectral image, spatiotemporal features are extracted.

[0092] In this step, the root cause of the defect is accurately located.

[0093] S44: Based on a lightweight machine learning model, the spatiotemporal features are quickly evaluated using hidden state vectors to obtain a compaction score.

[0094] S5: Train the subgrade multimodal data using a lightweight hybrid neural network and predict settlement, then detect the subgrade by subgrade settlement trend and compaction score.

[0095] To clarify the specific processing procedure for roadbed testing, step S5 includes S51 to S54, specifically:

[0096] S51: Extract multi-dimensional features from the roadbed multimodal data using a lightweight hybrid neural network and fuse them to obtain fused features;

[0097] To clarify the specific processing steps for the fusion features, step S5 includes S511 to S514, specifically:

[0098] S511: Based on the lightweight hybrid neural network, the roadbed multimodal data is processed, and temporal features are extracted by dynamically adjusting the number of neurons in the hidden layer;

[0099] In this step, the time dependence and dynamic response characteristics of time-series data such as the multi-directional vibration signals and pressure waveforms are captured based on the dynamic adjustment mechanism of the hidden layer neurons in the lightweight hybrid neural network. The dynamic adjustment mechanism of the hidden layer neurons dynamically adapts to the characteristics of different time-series data.

[0100] S512: Based on the combination of convolutional neural network and attention mechanism, spatial features are extracted by processing the multispectral image and preset sedimentation point cloud data;

[0101] In this step, spatial features are extracted by combining convolutional neural networks with attention mechanisms and processing the multispectral images and preset spatial data such as sedimentation point cloud data.

[0102] The convolutional neural network enhances the detection capability for minute defects such as cracks and settlement by reducing redundant computation. The introduction of an attention mechanism allows the model to focus more on key features, improving the efficiency and quality of feature extraction.

[0103] Preferably, the convolutional neural network is the EfficientNetV2 model.

[0104] S513: Based on the standardized transformation of the temperature and humidity data in the environmental characteristics, the environmental characteristics are obtained;

[0105] S514: Adaptively fuse the temporal features, spatial features, and environmental features based on a dynamic weighting mechanism to obtain fused features.

[0106] S52: Based on the fusion features and preset environmental compensation features, the compaction parameters of the road roller are adjusted through a feedback control algorithm to obtain optimized compaction parameters;

[0107] S53: Based on the optimized compaction parameters, train a time series model of the subgrade multimodal data to predict the subgrade settlement trend and obtain the settlement area;

[0108] In this step, a time series model of the subgrade multimodal data is trained based on the optimized compaction parameters. The trained time series model is used to predict the subgrade settlement trend and determine the settlement area.

[0109] The time series model can predict settlement trends in advance and detect potential problems in a timely manner.

[0110] S54: Based on the settlement area and compaction score, the subgrade is detected using a long short-term memory network.

[0111] Example 2:

[0112] This embodiment provides a rapid detection device for the effective hard layer of a high-fill roadbed, the device comprising:

[0113] The acquisition module is used to acquire roadbed multimodal data, which includes multi-directional vibration signals, environmental features, multispectral images, and pressure signals.

[0114] An alignment module is used to perform temporal and spatial alignment on the roadbed multimodal data to obtain aligned multimodal data;

[0115] The extraction module is used to extract features from the aligned multimodal data and obtain a multi-parameter fusion vector by performing unsupervised dimensionality reduction on the extracted features.

[0116] To clarify the specific methods for obtaining the extraction module, the following are included:

[0117] The acquisition unit is used to acquire real-time operating condition data;

[0118] An extraction unit is used to extract features from the aligned multimodal data to obtain an initial feature set;

[0119] To clarify the specific methods for obtaining the extraction units, the following are included:

[0120] The second unit is used to aggregate the pressure signal, generate grid-level pressure values ​​and perform feature calculations to obtain pressure features.

[0121] The decomposition unit is used to perform wavelet packet decomposition and frequency band quantization on the multi-directional vibration signal to obtain vibration characteristics.

[0122] The first calculation unit is used to calculate the mean and standard deviation of the pressure feature and the vibration feature based on the environmental features, respectively, to obtain the environmental correction factor;

[0123] The identification unit is used to analyze the multispectral image according to the target detection algorithm, identify roadbed cracks and calculate crack density to obtain image features;

[0124] An integration unit is used to integrate the environmental correction factor and the image features to obtain an initial feature set.

[0125] The first processing unit is used to perform dimensionality reduction processing on the high-dimensional features in the initial feature set according to the unsupervised machine learning model to obtain the dimensionality-reduced feature vector.

[0126] The adjustment unit is used to dynamically adjust the preset feature importance coefficients based on the real-time operating data to obtain dynamically weighted features;

[0127] The construction unit is used to construct a multi-parameter fusion vector based on the dynamically weighted features and the dimensionality-reduced feature vector.

[0128] The evaluation module is used to analyze the multi-parameter fusion vector in real time based on the edge nodes, and combine it with a lightweight machine learning model to quickly evaluate the compaction quality and obtain a compaction score.

[0129] To clarify the specific methods for obtaining the evaluation module, the following are included:

[0130] The second calculation unit is used to calculate the weights of the multi-parameter fusion vector based on the edge nodes and the online learning algorithm, and generate a dynamic weight vector.

[0131] The modeling unit is used to perform time-series modeling of the multi-directional vibration signal based on a long short-term memory network and extract the hidden state vector.

[0132] The detection unit is used to detect abnormal regions of the multi-directional vibration signal according to the dynamic weight vector, and to perform spatiotemporal alignment of the multispectral image through the abnormal regions to extract spatiotemporal features.

[0133] The evaluation unit is used to quickly evaluate spatiotemporal features based on a lightweight machine learning model and hidden state vectors to obtain a compaction score.

[0134] The prediction module is used to train the subgrade multimodal data based on a lightweight hybrid neural network and predict settlement, and to detect the subgrade through subgrade settlement trend and compaction score.

[0135] It should be noted that the specific manner in which each module performs its operation in the apparatus described in the above embodiments has been described in detail in the embodiments of the method, and will not be elaborated here.

[0136] Example 3:

[0137] Corresponding to the above method embodiments, this embodiment also provides a rapid detection device for the effective hard layer of a high embankment subgrade. The rapid detection device for the effective hard layer of a high embankment subgrade described below and the rapid detection method for the effective hard layer of a high embankment subgrade described above can be referred to in correspondence with each other.

[0138] Figure 2 This is a block diagram illustrating a rapid detection device 800 for the effective hard layer of a high-fill roadbed, according to an exemplary embodiment. Figure 2 As shown, the rapid detection device 800 for the effective hard layer of a high embankment subgrade may include: a processor 801 and a memory 802. The rapid detection device 800 for the effective hard layer of a high embankment subgrade may also include one or more of the following: a multimedia component 803, an I / O interface 804, and a communication component 805.

[0139] The processor 801 controls the overall operation of the rapid detection device 800 for the effective hard layer of a high embankment subgrade, thereby completing all or part of the steps in the aforementioned rapid detection method for the effective hard layer of a high embankment subgrade. The memory 802 stores various types of data to support the operation of the rapid detection device 800 for the effective hard layer of a high embankment subgrade. This data may include, for example, instructions for any application or method operating on the rapid detection device 800 for the effective hard layer of a high embankment subgrade, as well as application-related data, such as contact data, sent and received messages, images, audio, video, etc. The memory 802 can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in the memory 802 or transmitted via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. I / O interface 804 provides an interface between processor 801 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 805 is used for wired or wireless communication between the high-fill subgrade effective hard layer rapid detection device 800 and other devices. Wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 805 may include a Wi-Fi module, a Bluetooth module, and an NFC module.

[0140] In an exemplary embodiment, a rapid detection device 800 for the effective hard layer of a high embankment subgrade may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the aforementioned rapid detection method for the effective hard layer of a high embankment subgrade.

[0141] Example 4:

[0142] Corresponding to the above method embodiments, this embodiment also provides a medium. The medium described below can be referred to in conjunction with the rapid detection method for effective hard layer of high embankment subgrade described above.

[0143] A medium storing a computer program, which, when executed by a processor, implements the steps of a rapid detection method for the effective hard layer of a high embankment subgrade according to the above method embodiments.

[0144] The medium can specifically be any medium capable of storing program code, such as a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0145] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0146] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A rapid detection method for the effective hard layer of a high embankment subgrade, characterized in that, include: Acquire multimodal data of the roadbed, which includes multi-directional vibration signals, environmental features, multispectral images, and pressure signals; The roadbed multimodal data is time-space aligned to obtain aligned multimodal data; Feature extraction is performed on the aligned multimodal data, and unsupervised dimensionality reduction is applied to the extracted features to obtain a multi-parameter fusion vector; Based on real-time analysis of the multi-parameter fusion vector at the edge nodes, and combined with a lightweight machine learning model, the compaction quality is quickly evaluated to obtain a compaction score. The subgrade multimodal data is trained using a lightweight hybrid neural network to predict settlement, and the subgrade is detected by subgrade settlement trend and compaction score.

2. The rapid detection method for the effective hard layer of a high embankment subgrade according to claim 1, characterized in that, Feature extraction is performed on the aligned multimodal data, and unsupervised dimensionality reduction is applied to the extracted features to obtain a multi-parameter fusion vector, including: Obtain real-time operating data; Features are extracted from the aligned multimodal data to obtain an initial feature set; The high-dimensional features in the initial feature set are reduced in dimensionality using an unsupervised machine learning model to obtain the dimensionality-reduced feature vector. Based on the real-time operating data, the preset feature importance coefficients are dynamically adjusted to obtain dynamically weighted features; A multi-parameter fusion vector is constructed based on the dynamically weighted features and the dimensionality-reduced feature vector.

3. The rapid detection method for the effective hard layer of a high embankment subgrade according to claim 2, characterized in that, Features are extracted from the aligned multimodal data to obtain an initial feature set, including: The pressure signal is aggregated to generate grid-level pressure values ​​and feature calculations are performed to obtain pressure features; The multi-directional vibration signal is decomposed by wavelet packet and the frequency bands are quantized to obtain vibration characteristics; Based on the environmental characteristics, the mean and standard deviation of the pressure characteristics and the vibration characteristics are calculated respectively to obtain the environmental correction factor; The multispectral image is analyzed using a target detection algorithm to identify roadbed cracks and calculate crack density, thereby obtaining image features. An initial feature set is obtained by integrating the environmental correction factor and the image features.

4. The rapid detection method for the effective hard layer of a high embankment subgrade according to claim 1, characterized in that, Based on real-time analysis of the multi-parameter fusion vector at the edge nodes, and combined with a lightweight machine learning model, the compaction quality is rapidly evaluated to obtain a compaction score, including: The weights of the multi-parameter fusion vector are calculated based on the edge nodes and the online learning algorithm to generate a dynamic weight vector; The multi-directional vibration signal is time-series modeled based on a long short-term memory network, and the hidden state vector is extracted. Abnormal regions of the multi-directional vibration signal are detected based on the dynamic weight vector, and the multispectral image is spatiotemporally aligned using the abnormal regions to extract spatiotemporal features. Based on a lightweight machine learning model, spatiotemporal features are rapidly evaluated using hidden state vectors to obtain a compaction score.

5. The rapid detection method for the effective hard layer of a high embankment subgrade according to claim 1, characterized in that, The subgrade multimodal data is trained using a lightweight hybrid neural network to predict settlement. Subgrade detection is performed based on subgrade settlement trend and compaction score, including: Multidimensional features are extracted and fused from the roadbed multimodal data using a lightweight hybrid neural network to obtain fused features; Based on the integration of the fusion features and the preset environmental compensation features, the compaction parameters of the road roller are adjusted through a feedback control algorithm to obtain optimized compaction parameters; Based on the optimized compaction parameters, a time-series model of the subgrade multimodal data is trained to predict the subgrade settlement trend and obtain the settlement area; The subgrade is detected using a long short-term memory network based on the settlement area and compaction score.

6. The rapid detection method for the effective hard layer of a high embankment subgrade according to claim 5, characterized in that, Multidimensional features are extracted and fused from the roadbed multimodal data using a lightweight hybrid neural network to obtain fused features, including: The roadbed multimodal data is processed using a lightweight hybrid neural network, and temporal features are extracted by dynamically adjusting the number of neurons in the hidden layer. By combining convolutional neural networks with attention mechanisms, spatial features are extracted by processing the multispectral images and preset sedimentation point cloud data. The environmental characteristics are obtained by standardizing and transforming the temperature and humidity data in the environmental characteristics. The temporal features, spatial features, and environmental features are adaptively fused based on a dynamic weighting mechanism to obtain fused features.

7. A rapid detection device for the effective hard layer of a high embankment subgrade, characterized in that, include: The acquisition module is used to acquire roadbed multimodal data, which includes multi-directional vibration signals, environmental features, multispectral images, and pressure signals. An alignment module is used to perform temporal and spatial alignment on the roadbed multimodal data to obtain aligned multimodal data; The extraction module is used to extract features from the aligned multimodal data and obtain a multi-parameter fusion vector by performing unsupervised dimensionality reduction on the extracted features. The evaluation module is used to analyze the multi-parameter fusion vector in real time based on the edge nodes, and combine it with a lightweight machine learning model to quickly evaluate the compaction quality and obtain a compaction score. The prediction module is used to train the subgrade multimodal data based on a lightweight hybrid neural network and predict settlement, and to detect the subgrade through subgrade settlement trend and compaction score.

8. The rapid detection device for the effective hard layer of high embankment subgrade according to claim 7, characterized in that, The extraction module includes: The acquisition unit is used to acquire real-time operating condition data; The extraction unit is used to extract features from the aligned multimodal data to obtain an initial feature set; The first processing unit is used to perform dimensionality reduction processing on the high-dimensional features in the initial feature set according to the unsupervised machine learning model to obtain the dimensionality-reduced feature vector. The adjustment unit is used to dynamically adjust the preset feature importance coefficients based on the real-time operating data to obtain dynamically weighted features; The construction unit is used to construct a multi-parameter fusion vector based on the dynamically weighted features and the dimensionality-reduced feature vector.

9. The rapid detection device for the effective hard layer of high embankment subgrade according to claim 8, characterized in that, The extraction unit includes: The second unit is used to aggregate the pressure signal, generate grid-level pressure values ​​and perform feature calculations to obtain pressure features. The decomposition unit is used to perform wavelet packet decomposition and frequency band quantization on the multi-directional vibration signal to obtain vibration characteristics. The first calculation unit is used to calculate the mean and standard deviation of the pressure feature and the vibration feature based on the environmental features, respectively, to obtain the environmental correction factor; The identification unit is used to analyze the multispectral image according to the target detection algorithm, identify roadbed cracks and calculate crack density to obtain image features; An integration unit is used to integrate the environmental correction factor and the image features to obtain an initial feature set.

10. The rapid detection device for the effective hard layer of high embankment subgrade according to claim 7, characterized in that, The evaluation module includes: The second calculation unit is used to calculate the weights of the multi-parameter fusion vector based on the edge nodes and the online learning algorithm, and generate a dynamic weight vector. The modeling unit is used to perform time-series modeling of the multi-directional vibration signal based on a long short-term memory network and extract the hidden state vector. The detection unit is used to detect abnormal regions of the multi-directional vibration signal according to the dynamic weight vector, and to perform spatiotemporal alignment of the multispectral image through the abnormal regions to extract spatiotemporal features. The evaluation unit is used to quickly evaluate spatiotemporal features based on a lightweight machine learning model and hidden state vectors to obtain a compaction score.

Images