Construction impact identification method and system using geological radar technology

Abstract

The present invention provides a construction impact identification method and system using geological radar technology, including the following steps: collecting multi-frequency geological radar data and preprocessing and fusing them; converting the fused data into a depth domain and aligning it with a BIM model; extracting multi-scale features and performing classification and anomaly detection; building a spatiotemporal evolution model and generating a three-dimensional impact gradient grid. Adaptive signal processing technology is used to improve data quality, semi-supervised learning algorithms are used to achieve fine classification, and construction impacts are predicted through multi-scale analysis and spatiotemporal modeling. The accuracy and reliability of underground structure detection are improved, providing comprehensive and dynamic decision support for construction impact assessment in complex environments, and effectively balancing engineering construction and environmental protection needs.

Description

Technical Field

[0001] The present invention relates to geological radar technology, in particular to a construction impact identification method and system using geological radar technology. Background Art

[0002] With the acceleration of urbanization and the deepening of underground space development, the impact assessment of underground construction on the surrounding environment has become increasingly important. As a non-destructive and efficient underground detection method, geological radar technology plays a key role in the identification of construction impacts. Accurately assessing the impact of construction activities on underground structures and the surrounding environment is not only related to the safety and efficiency of the project, but also directly affects the stability of urban infrastructure and the quality of life of residents.

[0003] At present, the application of geological radar technology in construction impact identification has made some progress. Traditional methods mainly rely on single-frequency geological radar systems to evaluate underground structural changes through two-dimensional profile analysis. Some researchers have begun to try to use multi-frequency geological radar to improve detection accuracy, but data fusion technology is still relatively simple. At the same time, existing three-dimensional visualization technologies are mostly limited to the display of a single geological parameter and lack an organic combination with engineering design information. Some researchers have tried to combine geological radar data with other geophysical methods (such as resistivity tomography) to improve the reliability of interpretation. However, these methods still face challenges in processing data under complex geological conditions, especially in urban environments, where some areas have severe electromagnetic interference and complex underground structures.

[0004] Although some progress has been made in the research, there are still some specific problems in the use of geological radar technology for construction impact identification. For example, the collaborative processing of multi-frequency geological radar data is not good, especially when processing low signal-to-noise ratio signals, it is difficult to effectively extract deep weak reflections. For another example, under complex geological conditions, such as multi-layer soil structures or karst areas, the existing underground structure classification methods are not accurate enough, and it is difficult to accurately identify potential geological disaster points. For another example, the degree of integration between geological radar data and engineering design information is not high, and there is a lack of semantic-level integration, which makes it difficult to intuitively evaluate the impact of construction activities on specific engineering structures. In addition, in terms of spatiotemporal dynamic prediction, the existing models are difficult to accurately simulate geological changes during long-term construction, especially when considering the interaction of multiple geological factors. Finally, in terms of anomaly detection, the existing algorithms are not sensitive enough to small anomalies, and it is easy to ignore early signs of geological changes. The existence of these problems limits the application effect of geological radar technology in complex engineering environments and needs to be solved urgently. Summary of the invention

[0005] The purpose of the invention is to provide a construction impact identification method and system using geological radar technology, in order to solve one of the above-mentioned problems existing in the prior art.

[0006] Technical solution, according to one aspect of the present application, a construction impact identification method using geological radar technology includes the following steps:

[0007] Step S1, obtaining the original data set of each frequency collected by the multi-frequency geological radar system, and performing preprocessing including time zero point correction, direct wave elimination and gain correction; for the preprocessed data set, frequency domain adaptive filtering and space-time adaptive filtering are performed to eliminate interference from complex environments; multi-frequency data fusion is performed based on the filtered data set to obtain the final fused data; wherein the original data set of each frequency includes the original data of low frequency, medium frequency and high frequency;

[0008] Step S2, obtain the final fused data and perform time-depth conversion to generate depth domain data; use the interpolation algorithm to perform 3D volume reconstruction, obtain 3D GPR image data and import the BIM model data of the project, and perform spatial registration and alignment between the BIM model data and the 3D GPR image data; call the data fusion algorithm module to perform semantic-level fusion on the aligned GPR data and BIM data to generate a fused 3D model;

[0009] Step S3, obtaining the fused 3D model data and extracting a multi-scale feature set from it; using an unsupervised clustering algorithm to preliminarily classify the multi-scale feature set; calling a pre-configured semi-supervised learning algorithm module, and performing fine classification in combination with the labeled data; calling an anomaly detection algorithm module, and identifying potential abnormal areas based on the fine classification results;

[0010] Step S4, obtain the classification results and abnormal area information in step S3; extract multi-level features from micro to macro scales; call the cross-scale feature fusion network to integrate information at different scales; construct a spatiotemporal evolution model for the integrated multi-scale information, predict geological changes during the construction process, and obtain evolution results; design construction impact assessment indicators based on the evolution results; generate a three-dimensional impact gradient grid, visualize and output the impact range of different construction intensities.

[0011] According to one aspect of the present application, a construction impact identification system using geological radar technology includes:

[0012] at least one processor; and,

[0013] a memory communicatively connected to at least one of the processors; wherein,

[0014] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the construction impact identification method using geological radar technology as described in any of the above technical solutions.

[0015] Beneficial effects: This application realizes the effective integration of low, medium and high frequency geological radar data, improves the depth and resolution of underground structure detection, and enables the capture of comprehensive information from shallow details to deep structures. Combined with adaptive grid refinement and multi-resolution reconstruction technology, high-quality three-dimensional GPR images are generated and seamlessly integrated with BIM models to provide decision makers with an intuitive and information-rich visualization platform. Advanced machine learning algorithms, such as semi-supervised learning and anomaly detection, are used to achieve fine classification of underground structures and identification of risk points, improving the accuracy and reliability of geological assessments. By constructing a spatiotemporal evolution model and generating a three-dimensional impact gradient grid, dynamic prediction and visualization of construction impacts are achieved, providing a scientific basis for long-term risk assessment and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the present invention.

[0017] Figure 2 It is a flow chart of step S1 of the present invention.

[0018] Figure 3 It is a flow chart of step S2 of the present invention.

[0019] Figure 4 It is a flow chart of step S3 of the present invention.

[0020] Figure 5 It is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0021] like Figure 1 As shown in the figure, the construction impact identification method using geological radar technology includes the following steps:

[0022] Step S1, obtaining the original data set of each frequency collected by the multi-frequency geological radar system, and performing preprocessing including time zero point correction, direct wave elimination and gain correction; for the preprocessed data set, frequency domain adaptive filtering and space-time adaptive filtering are performed to eliminate interference from complex environments; multi-frequency data fusion is performed based on the filtered data set to obtain the final fused data; wherein the original data set of each frequency includes the original data of low frequency, medium frequency and high frequency;

[0023] Step S2, obtain the final fused data and perform time-depth conversion to generate depth domain data; use the interpolation algorithm to perform 3D volume reconstruction, obtain 3D GPR image data and import the BIM model data of the project, and perform spatial registration and alignment between the BIM model data and the 3D GPR image data; call the data fusion algorithm module to perform semantic-level fusion on the aligned GPR data and BIM data to generate a fused 3D model;

[0024] Step S3, obtaining the fused 3D model data and extracting a multi-scale feature set from it; using an unsupervised clustering algorithm to preliminarily classify the multi-scale feature set; calling a pre-configured semi-supervised learning algorithm module, and performing fine classification in combination with the labeled data; calling an anomaly detection algorithm module, and identifying potential abnormal areas based on the fine classification results;

[0025] Step S4, obtain the classification results and abnormal area information in step S3; extract multi-level features from micro to macro scales; call the cross-scale feature fusion network to integrate information at different scales; construct a spatiotemporal evolution model for the integrated multi-scale information, predict geological changes during the construction process, and obtain evolution results; design construction impact assessment indicators based on the evolution results; generate a three-dimensional impact gradient grid, visualize and output the impact range of different construction intensities.

[0026] In this embodiment, the acquisition and processing of multi-frequency geological radar data lays a high-quality data foundation for subsequent analysis. Through adaptive filtering and multi-frequency data fusion, the signal-to-noise ratio and resolution of the data are improved, so that even weak geological features can be accurately captured. It is of great significance for identifying potential geological risk points, such as weak interlayers or caves. The fusion of 3D GPR image reconstruction and BIM model realizes the high integration of underground geological information and engineering design information, which not only improves the accuracy of spatial positioning, but also provides engineers with a unified and information-rich decision-making platform. For example, in the development of urban underground space, the integrated model can intuitively display the spatial relationship between underground pipelines, geological structures and design structures, reducing the risk of construction conflicts. Multi-scale feature extraction and anomaly detection realize the fine classification and risk identification of underground structures through advanced machine learning algorithms. It can not only adapt to complex geological environments, but also effectively utilize limited labeled data to improve the accuracy and reliability of classification. In long-term construction monitoring, adaptive learning ability is particularly important, and the model can be continuously updated to reflect the dynamic changes of geological conditions. Multi-scale analysis and spatiotemporal evolution modeling transform static geological information into dynamic impact prediction. By constructing a spatiotemporal evolution model and generating a three-dimensional impact gradient grid, the long-term impact of construction activities on the underground environment can be predicted. For example, in subway tunnel construction, the prediction model can evaluate the impact of long-term excavation on groundwater flow fields and surface subsidence, providing a scientific basis for the formulation of sustainable construction strategies. In short, it not only improves the safety and efficiency of underground projects, but also provides important technical support for urban sustainable development and rational use of underground space, making complex underground project decisions more scientific and controllable, and effectively balancing the multiple needs of engineering construction, environmental protection and economic benefits.

[0027] It should be noted that, in this application, in order to make the description smoother, each frequency refers to low frequency, medium frequency and high frequency.

[0028] like Figure 2 As shown, the step S1 is specifically as follows:

[0029] Step S11, using a multi-frequency geological radar system to simultaneously collect data of each frequency, including low-frequency, medium-frequency and high-frequency data; on a preset survey line, using antennas of different frequencies to collect data; setting a sampling interval and time window for each frequency; recording the two-way travel time and signal amplitude of each sampling point; storing the collected low-frequency, medium-frequency and high-frequency data as original data sets, including the number of sampling points, sampling interval and time window;

[0030] Step S12, obtaining the original data set; performing time zero point correction on the data set of each frequency to ensure that the signal start time is consistent; calculating the wavelength and sampling interval corresponding to each frequency, and determining the adaptive window size; using the median filtering method of the adaptive window size to process the low-frequency, medium-frequency and high-frequency data, and remove the direct wave; calling the constructed adaptive gain function containing adjustable parameters, and applying it to the data after removing the direct wave, to obtain the data set after gain correction;

[0031] Step S13, obtaining a data set after gain correction; performing a fast Fourier transform on the data set of each frequency to obtain frequency domain data; calling an adaptive notch filter including adjustable interference frequency and bandwidth parameters, applying it to the frequency domain data, and performing a filtering operation; performing an inverse fast Fourier transform on the filtered frequency domain data to obtain time domain data; calling a four-dimensional filter including a spatial weight function and a time weight function and applying it to the time domain data to obtain each frequency data after removing the interference of the complex environment;

[0032] Step S14, obtaining each frequency data after removing the interference of complex environment; performing multi-scale wavelet decomposition on the data set of each frequency to obtain respective approximate coefficients and detail coefficients; calling the constructed adaptive weight function considering frequency and depth factors; and using the adaptive weight function to fuse the approximate coefficients and detail coefficients of different frequency data; using the fused approximate coefficients and detail coefficients to perform inverse wavelet transform to obtain the final multi-frequency fusion data.

[0033] In this embodiment, low-frequency, medium-frequency and high-frequency data are collected simultaneously, making full use of the complementary advantages of different frequency signals. Low-frequency signals have a large penetration depth and are suitable for detecting deeper geological structures; medium-frequency signals balance depth and resolution; high-frequency signals have high resolution and are suitable for detecting shallow details. The multi-frequency acquisition strategy improves the comprehensiveness and reliability of the data. The use of time zero correction ensures the time consistency of data of different frequencies, laying the foundation for subsequent fusion. Direct wave elimination and gain correction further improve the data quality and enhance the identifiability of deep signals. Frequency domain adaptive filtering and space-time adaptive filtering effectively eliminate complex environmental interference, such as electromagnetic noise and ground clutter, and improve the signal-to-noise ratio of the signal. Multi-frequency data fusion fully integrates the advantageous information of each frequency data through wavelet decomposition and adaptive weight fusion, generating a high-quality data set containing rich underground information. In short, not only the data quality is improved, but also the multi-scale information is retained, providing a reliable data basis for subsequent geological structure identification and change monitoring. In construction impact assessment, this high-quality, multi-scale data can more accurately reflect subtle changes in underground structures, which is conducive to the timely discovery of potential construction risks such as ground subsidence and cavity formation, thereby providing an important basis for construction decision-making.

[0034] like Figure 3 As shown, step S2 is specifically as follows:

[0035] Step S21, obtaining the final multi-frequency fusion data; calling the adaptive speed function considering the spatial position factor to perform time-depth conversion on the fusion data to generate depth domain data; using the improved interpolation algorithm module considering the spatial correlation to perform 3D volume reconstruction on the depth domain data to generate high-resolution 3D GPR image data;

[0036] Step S22, obtaining the BIM model data of the project; and importing the 3D GPR image data; calling the spatial registration algorithm module including the feature point extraction and matching module to accurately align the 3D GPR image data and the BIM model data; generating the aligned GPR data set and BIM data set, maintaining the integrity and spatial consistency of the original data;

[0037] Step S23, obtaining the aligned GPR data set and BIM data set; calling the data fusion algorithm module including feature extraction, matching and fusion modules to extract geological features from the GPR data, and extracting geometric and semantic features from the BIM data; calling the fusion function that considers the weights and interaction relationships of data from different sources to process the extracted features, and generating a fused 3D model including geological information and building information.

[0038] In this embodiment, the time-depth conversion is performed by an adaptive velocity function that considers the spatial position factor, which overcomes the depth error problem caused by the inaccurate velocity model in the traditional method. The adaptive method can dynamically adjust the velocity model according to the geological characteristics of different regions, thereby improving the accuracy of the depth conversion. An improved interpolation algorithm is used for 3D volume reconstruction to generate a high-resolution 3D GPR image. Taking into account the spatial correlation, it is possible to effectively fill the data blank area while maintaining the original data characteristics, providing a more continuous and complete underground structure representation. The spatial registration and alignment of the BIM model and the 3D GPR image realizes the precise fusion of the two heterogeneous data through feature point extraction and matching. Not only the integrity of the original data is maintained, but also the spatial association between the underground geological structure and the ground building structure is established. Through the semantic-level fusion algorithm, the geological information and the building information are organically combined to generate a comprehensive 3D model containing rich semantic information. The fusion model provides unprecedented information and visualization capabilities for construction impact assessment. In practical applications, the highly integrated 3D model can intuitively display key information such as geological conditions, underground pipelines, and building foundations, enabling engineers to comprehensively evaluate the potential impact of construction activities on the underground environment and ground buildings. For example, it is possible to clearly see the underground structures that may be affected by excavation work, or predict the impact of foundation settlement on surrounding buildings. This improves the accuracy of risk assessment and the scientific nature of decision-making, and helps to formulate more accurate construction plans and protection measures.

[0039] like Figure 4 As shown, the step S3 is specifically as follows:

[0040] Step S31, obtaining a fused 3D model; using a pre-configured multi-scale feature extractor including spatial, frequency and texture features to process the 3D model data to generate a multi-dimensional feature vector; normalizing and reducing the dimensionality of the extracted multi-dimensional feature vector to obtain a final feature set;

[0041] Step S32, obtaining a final feature set; using a density peak clustering algorithm module including an adaptive density estimation module and a relative density difference calculation module to perform cluster analysis on the final feature set to generate a preliminary classification result; performing a post-processing step to merge similar clusters and eliminate isolated points to obtain an optimized clustering result;

[0042] Step S33, obtain preliminary classification results and a small amount of manually labeled sample data; build a graph model and calculate the similarity between samples; call the semi-supervised learning algorithm module including the label propagation and manifold regularization modules to refine the preliminary classification results to obtain the final classification results; based on the final classification results, calculate the anomaly score of each sample point; and identify potential abnormal areas according to the set adaptive threshold, and output the abnormal area set.

[0043] In this embodiment, multi-scale feature extraction fully captures the multi-level information of underground structures, including features such as texture, shape, and spatial distribution. It can focus on large-scale geological units and small-scale detail changes at the same time, which improves the comprehensiveness and accuracy of subsequent analysis. The application of unsupervised clustering algorithms realizes the preliminary classification of complex underground structures, does not rely on prior knowledge, and can adaptively discover the intrinsic structure in the data, which is particularly suitable for dealing with complex geological environments and insufficient prior information. The introduction of semi-supervised learning algorithms further improves the accuracy of classification. By combining a small amount of labeled data and a large amount of unlabeled data, it is possible to use expert knowledge to improve the accuracy and interpretability of classification while maintaining flexibility. The application of anomaly detection algorithms is crucial to identifying potential geological risks. By analyzing the statistical characteristics and spatial distribution of data, areas that are significantly different from the surrounding environment, such as cavities, faults, or abnormal aquifers, can be effectively identified. Fine classification results can help engineers accurately identify different types of geological bodies, such as soft strata, karst areas, or aquifers, so as to specifically evaluate the impact of construction on different geological units. For example, in tunnel construction, vulnerable areas that may cause water inrush or collapse can be accurately located. The anomaly detection function can timely detect potential geological disaster risks, such as the expansion of underground cavities or deformation of strata caused by construction. Early warning capabilities are of great significance for preventing construction risks and ensuring project safety. In addition, the model can be continuously updated and optimized as the construction progresses, realizing dynamic monitoring and evaluation of changes in geological conditions, and providing continuous technical support for long-term construction safety management.

[0044] like Figure 5 As shown, step S4 is specifically as follows:

[0045] Step S41, obtaining the final classification results and abnormal area information; extracting features from three scales, micro, meso and macro, including texture, structure and geological unit features; inputting the extracted features into the cross-scale feature fusion network of the attention mechanism, realizing feature fusion through adaptive weights; outputting the fused multi-scale features;

[0046] Step S42, obtaining the fused multi-scale features; using a numerical method to solve the spatiotemporal evolution equation including the diffusion term, the reaction term and the attenuation term to obtain the evolution state at different time points; interpolating and smoothing the evolution results to generate continuous spatiotemporal evolution data;

[0047] Step S43, obtaining spatiotemporal evolution data; calculating the impact index value of each spatial point at different times according to the preset construction impact assessment index; mapping the index value to different impact levels according to the set impact level classification standard; generating a three-dimensional impact index field; wherein the construction impact assessment index includes a state change index, a spatial gradient index and a curvature index; and the three-dimensional impact index field includes spatial coordinates and corresponding impact index values;

[0048] Step S44, obtaining a three-dimensional impact index field; using a three-dimensional isosurface generation algorithm module including an adaptive mesh refinement module to generate isosurfaces of different impact levels; constructing a three-dimensional impact gradient mesh; superimposing the generated isosurfaces and gradient meshes with the BIM model data; and outputting a final three-dimensional construction impact visualization model;

[0049] Among them, each grid cell in the three-dimensional impact gradient grid contains location coordinates and impact level information; the three-dimensional construction impact visualization model contains underground structure, impact isosurface and color-coded impact gradient grid.

[0050] In this embodiment, multi-level feature extraction from micro to macro captures all-round information of geological structure. It can simultaneously focus on micro crack development and macro stratum deformation, providing a comprehensive basis for geological change assessment. The cross-scale feature fusion network effectively integrates information of different scales through an adaptive weight mechanism, overcomes the problem of scale isolation in traditional methods, and realizes seamless connection from micro structure to macro geological units. In the spatiotemporal evolution model, by solving the spatiotemporal evolution equation containing diffusion, reaction and attenuation terms, it is possible to simulate and predict the dynamic geological changes during the construction process. It is crucial to evaluate the long-term construction impact and can foresee potential geological risks, such as the acceleration of stratum subsidence rate or changes in groundwater flow patterns. The design of the construction impact assessment index takes into account multiple dimensions such as state change, spatial gradient and curvature, fully reflects the complexity of geological changes, can capture subtle changes that may be ignored by a single indicator, and improves the accuracy and reliability of the assessment. The generation and visualization of the three-dimensional impact gradient grid can present the spatial distribution of construction impacts with high precision through adaptive grid refinement and isosurface generation algorithms. The 3D visualization model intuitively shows the scope of influence under different construction intensities, providing decision makers with a clear and intuitive reference. In practical applications, for example, in subway tunnel construction, it can predict the scope of influence of excavation on surrounding strata, identify areas where excessive deformation may occur, and dynamically adjust construction parameters to minimize adverse effects. For large foundation pit projects, it can monitor the stress state of the support structure and the deformation trend of the surrounding strata in real time, and promptly detect potential instability risks. In addition, it can also evaluate the impact of construction on the groundwater system, predict possible ground subsidence or groundwater pollution problems, and provide a scientific basis for environmental protection and sustainable construction. In general, this step realizes the full process integration from data analysis to decision support, and improves the scientificity, accuracy and operability of construction impact assessment.

[0051] According to one aspect of the present application, in step S12, a median filtering method with an adaptive window size is used to process the low-frequency, medium-frequency and high-frequency data to remove the direct wave, specifically:

[0052] Step S121, obtaining low-frequency, medium-frequency and high-frequency data sets after time zero point correction;

[0053] Step S122: construct an adaptive structural element according to the signal characteristics of each frequency, and the shape parameters are dynamically adjusted according to the signal frequency;

[0054] Step S123, applying an opening operation to each frequency data set to remove the peaks in the data; applying a closing operation to the result of the opening operation to fill the valleys in the data; adding and averaging the results of the opening operation and the closing operation to obtain an estimated background signal;

[0055] Step S124, calculating the difference between the original data and the estimated background signal to obtain a residual signal after removing the direct wave; outputting the processed low-frequency, medium-frequency and high-frequency data sets with the direct wave component removed.

[0056] In this embodiment, a morphological filtering method of adaptive structural elements is adopted. It can adapt to the characteristics of data of different frequencies by dynamically adjusting the shape parameters of structural elements according to the characteristics of signals of different frequencies, so as to achieve more accurate background signal estimation. The limitations of the traditional fixed window filtering method are overcome by adaptability, and the interference of direct waves can be effectively removed while retaining useful information. The combined application of open operation and closed operation cleverly solves the problem of peaks and valleys in the data. The removal of peaks by open operation helps to eliminate high-frequency noise and abnormal signals, while the filling of valleys by closed operation helps to restore weak signals that may be deleted by mistake. The combined operation not only improves the accuracy of direct wave estimation, but also maintains the integrity of the original signal. The residual signal is obtained by calculating the difference between the original data and the estimated background signal, which can effectively separate the direct wave component and the underground reflection signal. In the application of construction impact identification, high-quality de-direct wave processing makes the reflection signal of shallow geological structure more clear and discernible, which is crucial for evaluating the impact of construction on the soil layer near the surface. For example, in shallow foundation construction, soil structure changes or potential pipeline locations can be more accurately identified to avoid accidental damage during construction. For deep structures, such as tunnels or deep foundation pits, the data after removing the direct wave can provide clearer deep geological interface information, which helps to evaluate the impact of construction on deep geological structures. It lays the foundation for the subsequent multi-frequency data fusion, so that the fused data can more comprehensively and accurately reflect the complexity of the underground environment. In general, this step enhances the application value of geological radar technology in construction impact assessment by improving data quality, and provides solid data support for more accurate and reliable construction decisions.

[0057] According to one aspect of the present application, in step S13, an adaptive notch filter is used to perform a filtering operation, specifically:

[0058] Step S131, obtaining transformed low-frequency, medium-frequency and high-frequency data sets; performing wavelet packet transform on each data set; using a predetermined wavelet basis function and decomposition layer number, performing wavelet packet decomposition to obtain a time-frequency representation of each frequency data; outputting wavelet packet coefficients of low-frequency, medium-frequency and high-frequency data;

[0059] Step S132, obtaining the wavelet packet coefficients of each frequency data set; calculating the noise standard deviation of each decomposition level and node; defining an adaptive threshold for each wavelet packet coefficient according to the signal length and the noise standard deviation; performing threshold processing on the wavelet packet coefficient according to the designed soft threshold function; and outputting the denoised wavelet packet coefficients;

[0060] Step S133, obtain the denoised wavelet packet coefficients; select the same wavelet basis function as that used in the decomposition; perform inverse wavelet packet transform to reconstruct the time domain signal; obtain filtered data of low frequency, medium frequency and high frequency respectively; and output the filtered low frequency, medium frequency and high frequency data sets.

[0061] In this embodiment, through advanced signal processing technology, efficient and accurate noise removal and signal enhancement are achieved, and the quality and interpretability of geological radar data are improved. Compared with the traditional Fourier transform, the wavelet packet transform can provide localized analysis in both the time and frequency domains, and is more suitable for processing non-stationary geological radar signals. It can accurately capture the transient characteristics and local changes in the signal, providing a more detailed time-frequency representation for subsequent noise removal. The design and application of adaptive thresholds are another improvement point in this step. By considering the signal length and the standard deviation of noise, an adaptive threshold is defined for each wavelet packet coefficient, overcoming the limitations of the traditional fixed threshold method. This adaptive method can flexibly adjust the denoising intensity at different scales and frequencies, effectively balancing noise suppression and signal retention. The use of soft threshold functions further improves the smoothness of denoising and avoids the discontinuity that may be caused by hard thresholds. The time domain signal is reconstructed by inverse wavelet packet transform to ensure the integrity and coherence of the processed data. In the application of construction impact identification, high-quality noise removal makes weak geological reflection signals more clearly discernible, which is crucial for evaluating the subtle impact of construction on underground structures. For example, in tunnel construction, it is possible to more accurately identify rock interfaces or fault zones and predict possible water inrush or collapse risks. For foundation pit projects, denoised data can provide more accurate soil layer structure information, which helps to optimize support design and assess the settlement risk of surrounding buildings. It lays the foundation for subsequent multi-frequency data fusion, so that the fused data can more comprehensively and accurately reflect the complexity of the underground environment. Especially in urban environments, where underground structures are complex and electromagnetic interference is severe in some places, denoising technology can improve the reliability and resolution of geological radar detection. In general, this step enhances the application value of geological radar technology in construction impact assessment by improving data quality and signal-to-noise ratio, and provides solid data support for more accurate and reliable construction decisions and risk assessments.

[0062] According to one aspect of the present application, in step S14, multi-scale wavelet decomposition is performed on the data set of each frequency to obtain respective approximate coefficients and detail coefficients, specifically:

[0063] Step S141, obtaining each frequency data after removing the interference of the complex environment, performing empirical mode decomposition on the data set of each frequency; iteratively extracting each intrinsic mode function until the remaining signal becomes a monotonic function; recording the intrinsic mode function set of each frequency data; outputting the intrinsic mode function set of low-frequency, medium-frequency and high-frequency data;

[0064] Step S142, obtaining the intrinsic mode function set of each frequency; selecting a predetermined wavelet function; performing continuous wavelet transform on each intrinsic mode function; calculating wavelet coefficients of different scales and time positions; outputting the continuous wavelet transform results of all intrinsic mode functions of each frequency data;

[0065] Step S143, obtain the continuous wavelet transform result; calculate the energy distribution of each frequency signal at different scales; design an adaptive weight function according to the energy distribution; use the weight function to fuse the wavelet transform results of different frequencies; perform an inverse transform on the fused wavelet transform result to separate the approximate coefficients and the detail coefficients; output the fused time-frequency representation, as well as the corresponding approximate coefficients and detail coefficients.

[0066] In this embodiment, the empirical mode decomposition (EMD) can adaptively decompose complex signals into a series of intrinsic mode functions (IMFs), which is particularly suitable for processing nonlinear and non-stationary geological radar signals. Through EMD, geological features of different scales, such as shallow reflections, deep interfaces, etc., can be effectively separated, providing an ideal basis for subsequent multi-frequency fusion. Performing continuous wavelet transform on each IMF is another improvement. Continuous wavelet transform provides a time-frequency representation of the signal, which can accurately capture the local characteristics and transient changes of the signal. Compared with the traditional Fourier transform, it can better retain the time information of the signal, which helps to identify the subtle changes of the underground structure. The adaptive weight fusion strategy based on energy distribution is the core improvement of this step. By analyzing the energy distribution of different frequency signals at each scale and designing an adaptive weight function, the optimal fusion of low-frequency, medium-frequency and high-frequency data is achieved. The complementary advantages of different frequency signals are fully utilized: low-frequency data provides deep information, high-frequency data provides high-resolution shallow details, and medium-frequency data balances depth and resolution. The effective fusion of multi-frequency data makes the detection of underground structures more comprehensive and accurate. For example, in subway tunnel construction, the fused data can simultaneously provide deep geological structure information and shallow soil deformation details, which helps to comprehensively assess the impact of excavation on the surrounding environment. For large foundation pit projects, multi-scale information can help engineers more accurately assess the stress state of the support structure and the deformation trend of the surrounding strata, and promptly detect potential instability risks. In addition, under complex geological conditions, such as karst areas or multi-layer soil structures, fusion technology can provide richer geological information and help identify potential geological disaster risk points. In general, this step improves the detection capability and reliability of geological radar technology in complex environments by efficiently integrating multi-frequency data, providing more comprehensive and accurate underground environmental information for construction impact assessment, thereby supporting more scientific construction decisions and risk management.

[0067] According to one aspect of the present application, in step S21, the final multi-frequency fusion data is obtained; the adaptive speed function considering the spatial position factor is called to perform time-depth conversion on the fusion data to generate depth domain data, specifically:

[0068] Step S211, obtaining the final multi-frequency fusion data, using the dynamic time warping algorithm to calculate the similarity between local waveforms; constructing a similarity matrix; applying the spectral clustering algorithm to group the waveforms based on the similarity matrix; and outputting the waveform clustering results;

[0069] Step S212, obtaining waveform clustering results; for each cluster, using a similarity analysis method to estimate the speed; calculating waveform correction and similarity values ​​at different speeds; selecting the speed with the largest similarity value as the estimated speed of the cluster; constructing an initial speed field, including spatial positions and corresponding speed estimates;

[0070] Step S213, obtain the initial velocity field; apply the tensor voting framework to smooth and extrapolate the velocity field; design the objective function, including the travel time fitting term and the regularization term; use the L-BFGS algorithm to optimize the velocity field and minimize the objective function; obtain the optimized three-dimensional velocity field; use the optimized velocity field to perform time-depth conversion on the fused data of step S14; output the depth domain data.

[0071] In this embodiment, DTW can effectively calculate the similarity between local waveforms, overcoming the disadvantage of the traditional correlation analysis method being insensitive to time axis expansion. It is particularly suitable for processing the common time delay and waveform deformation problems in geological radar signals, and provides a reliable similarity measure for subsequent velocity estimation. The introduction of the spectral clustering algorithm realizes the automatic grouping of complex waveforms. This unsupervised learning method can adaptively discover the intrinsic structure in the data, and is particularly suitable for processing situations with complex geological environments and diverse waveforms. Through waveform clustering, the electromagnetic wave propagation speed of different geological units can be more accurately estimated. The application of the tensor voting framework and the L-BFGS optimization algorithm is another improvement of this step. The tensor voting framework can effectively use spatial correlation to smooth and extrapolate the velocity field, overcoming the problem of discontinuous velocity estimation in traditional methods. The L-BFGS algorithm achieves global optimization of the velocity field by optimizing the objective function while ensuring the travel time fitting accuracy. High-precision time-depth conversion makes the spatial positioning of underground structures more accurate, which is crucial for evaluating the impact of construction on specific depth layers. For example, in tunnel construction, being able to more accurately locate the location of surrounding rock fracture zones or aquifers can help optimize support design and predict water inrush risks. For deep foundation pit projects, accurate depth information can help engineers better assess the impact of excavation on soil layers at different depths and optimize layered excavation and support plans. In addition, in underground pipeline detection, high-precision depth positioning can reduce the risk of accidental damage to pipelines during construction. In general, by improving the spatial positioning accuracy of underground structures, the application value of geological radar technology in construction impact assessment is enhanced, providing key technical support for more accurate and reliable construction decisions and risk assessments.

[0072] According to one aspect of the present application, in step S21, an improved interpolation algorithm module that considers spatial correlation is used to perform 3D volume reconstruction on the depth domain data to generate high-resolution 3D GPR image data, specifically:

[0073] Step S214, obtain depth domain data and optimized velocity field. Define anisotropic radial basis functions, considering spatial anisotropy. Construct interpolation functions, including weight coefficients and basis functions. Solve the linear equations to determine the weight coefficients. Apply the interpolation function to the entire three-dimensional space to obtain preliminary interpolation results.

[0074] Step S215, obtain preliminary interpolation results. Initialize a uniform coarse grid. Calculate the local interpolation error of each grid cell. Set an error threshold and refine the grid in areas exceeding the threshold. Iterate the refinement process until the accuracy requirement is met or the maximum number of refinements is reached. Output the refined multi-resolution grid.

[0075] Step S216, obtain a multi-resolution grid. Perform radial basis function interpolation on the coarse grid and the refined grid respectively. Design an adaptive weight function based on the local interpolation error. Use the weight function to fuse the interpolation results of different resolutions. Generate the final 3D GPR image data, including high-precision areas and smooth transition areas.

[0076] In this embodiment, compared with the traditional isotropic function, the anisotropic RBF can better adapt to the spatial variation characteristics of the geological structure, especially in the geological environment with obvious layered structures or faults and other heterogeneous features, and can provide more accurate interpolation results. Not only the spatial resolution of the reconstructed image is improved, but also the detailed features of the geological structure are better preserved. The use of adaptive multi-resolution grids is another improvement. By dynamically adjusting the grid density according to the local interpolation error, efficient use of computing resources is achieved. In areas with complex geological structures or drastic signal changes, the grid will be automatically refined to capture more details; while in relatively uniform areas, a coarser grid is maintained to save computing resources. This adaptive strategy not only improves the reconstruction efficiency, but also achieves accurate expression of important features. The adaptive weight fusion method based on local interpolation error effectively integrates the interpolation results on grids of different resolutions. It can achieve smooth transition while ensuring high accuracy, avoiding the common seam problem in traditional multi-resolution methods. High-resolution and high-precision 3D GPR images provide strong support for the fine characterization of underground structures. For example, in the development of urban underground space, this fine 3D model can accurately present the complex distribution of underground pipe networks and reduce the risk of pipeline damage during construction. For large-scale underground projects, such as subway stations or underground shopping malls, high-precision 3D models can help engineers accurately assess the impact of excavation on surrounding strata and groundwater and optimize construction plans. In addition, in complex geological conditions such as karst areas or soft soil foundations, this high-resolution model can clearly display potential geological disaster risks, such as caves or weak interlayers, providing key basis for risk assessment and preventive measures. In general, this step enhances the visualization and spatial cognition of the underground environment by generating high-quality 3D GPR images, providing more comprehensive and intuitive information support for construction impact assessment, thereby helping to formulate more accurate and safe construction strategies.

[0077] According to one aspect of the present application, step S23 is specifically:

[0078] Step S231, obtain aligned 3D GPR image data and BIM model data. Extract SURF feature points from GPR data, including positions and descriptors. Extract geometric feature points, such as corners and edges, from the BIM model. Use a fast approximate nearest neighbor algorithm to match based on feature descriptors. Output matched feature point pairs.

[0079] Step S232, obtain matching feature point pairs. Construct a factor graph, where vertices represent feature points and edges represent relationships between feature points. Define a data consistency factor to measure the matching degree between GPR and BIM data. Define a spatial smoothing factor to maintain local consistency. Construct an energy function to comprehensively consider all factors. Output the constructed probability graph model.

[0080] Step S233, obtain the probability graph model. Initialize the message and confidence. Iterate the message passing process to update the message and confidence of each node. Check the convergence condition, and continue to iterate if it has not converged. After convergence, determine the optimal state of each node according to the final confidence. Fuse the GPR and BIM data based on the optimization results to generate a fused 3D model.

[0081] In this embodiment, the application of SURF feature extraction and fast approximate nearest neighbor algorithm is an important improvement. SURF features are scale and rotation invariant, and can effectively capture key points in GPR images and BIM models, while the fast approximate nearest neighbor algorithm improves the efficiency of feature matching. This combination not only realizes the rapid alignment of two heterogeneous data, but also ensures the accuracy of alignment, especially in complex underground environments, the effect is better. The construction of factor graphs and the design of energy functions are another improvement. By modeling the relationship between feature points as a probabilistic graph model, data consistency and spatial smoothness can be comprehensively considered. This global optimization strategy not only improves the robustness of fusion, but also effectively handles local matching errors or data missing problems. The application of message passing algorithm realizes efficient model solving. By iteratively updating the messages and confidence between nodes, it is possible to effectively process large-scale data sets while ensuring global optimality. The semantic-level fusion of GPR data and BIM models provides strong support for a comprehensive understanding of the underground environment. For example, in the construction of urban underground pipe corridors, the fusion model can intuitively display the spatial relationship between underground pipelines and designed pipe corridors, which helps to optimize wiring and avoid conflicts. For deep foundation pit projects, this fusion model can accurately display the relative position of the excavation surface and the foundations of surrounding buildings, providing a key basis for evaluating the impact of construction on neighboring buildings. In addition, in the construction of historical and cultural heritage protection areas, the fusion model can simultaneously display the distribution of underground cultural relics and the above-ground building structures, which is helpful for formulating construction plans that both protect cultural relics and ensure the safety of the project. In general, this step improves the comprehensiveness and accuracy of construction impact assessment by achieving a high degree of integration of GPR data and BIM models. It provides engineers with an intelligent decision-making platform that integrates geological information, engineering design, and construction simulation, which helps to better predict and manage various risks in the construction process, thereby improving the safety and efficiency of the project.

[0082] According to one aspect of the present application, step S32 is specifically:

[0083] Step S321, obtain a feature data set. Calculate the distance matrix between all data points in the feature space. Design a kernel function with adaptive bandwidth. For each data point, use the kernel function to calculate its local density. Consider the unevenness of data distribution and dynamically adjust the bandwidth parameter. Output the local density estimate of each data point.

[0084] Step S322, obtain local density estimates and distance matrices. For each data point, find the nearest neighbor with higher density. Calculate the distance to the nearest neighbor as the relative density difference. For the point with the highest local density, specify its relative density difference as the maximum distance. Calculate the cluster center score for each point, taking into account the local density and relative density difference. Output the relative density difference and cluster center score for each point.

[0085] Step S323, obtain the cluster center score. Select the point with a higher score as the initial cluster center. Use the agglomerative hierarchical clustering method to gradually merge the neighbor clusters. Define and calculate the distance between clusters, and select the cluster pair with the minimum distance to merge. Set a distance threshold, and stop merging when the minimum distance between clusters is greater than the threshold. Output the final clustering result, that is, the optimized clustering result, including the cluster center and the cluster label of each data point.

[0086] In this embodiment, by dynamically adjusting the bandwidth parameters, it is possible to adapt to the unevenness of data distribution, use a larger bandwidth in sparse areas to increase statistical reliability, and use a smaller bandwidth in dense areas to improve local accuracy. This adaptive strategy improves the accuracy of density estimation, especially in complex geological environments. The calculation method of relative density difference is another improvement. By finding the nearest neighbor with higher density and calculating the distance difference, the cluster center and outliers can be effectively identified. It not only overcomes the problem of parameter sensitivity of traditional density clustering algorithms, but also adaptively discovers cluster structures of different scales. The application of agglomerative hierarchical clustering achieves the optimization of clustering results. By gradually merging neighbor clusters and setting a distance threshold to control the merging process, it is possible to eliminate the noise effect caused by over-segmentation while maintaining important geological features. High-precision classification of geological units provides a key basis for evaluating the impact of construction on different types of strata. For example, in tunnel construction, accurate identification of lithology distribution and fault location can help engineers predict possible water inrush or rock burst risks and optimize support design. For large foundation pit projects, accurate soil layer classification helps to evaluate the stability of soil at different depths and locations and guide the formulation of layered excavation and support plans. In addition, in construction in groundwater-rich areas, the distribution of aquifers and aquicludes can be effectively identified, providing an important reference for assessing the impact of construction on the groundwater system. In general, this step enhances the pertinence and accuracy of construction impact assessment by providing high-quality geological unit classification results. It enables engineers to develop more refined construction plans and risk prevention and control measures based on the characteristics of different geological units, thereby improving the safety and economy of the project.

[0087] According to one aspect of the present application, step S33 is specifically:

[0088] Step S331, obtain the optimized initial clustering results and labeled data. Construct a k-nearest neighbor graph and determine the k nearest neighbors of each data point. Calculate the similarity between neighbors as the weight of the edge. Construct a weight matrix and a degree matrix. Calculate the normalized Laplace matrix for subsequent manifold regularization. Output the constructed graph structure, including the vertex set, edge set and Laplace matrix.

[0089] Step S332, obtain the graph structure and labeled data. Design the objective function, including the fitting term of the labeled data, the graph regularization term and the function smoothing term. Use gradient descent or conjugate gradient method to solve the optimization problem. Iterate and update the classification function until convergence or the maximum number of iterations is reached. Output the optimized classification function and assign category information to each data point.

[0090] Step S333, obtain the optimized classification function. Calculate the deviation of each data point from its marked (or predicted) category. Consider the category information of neighboring points and calculate the local consistency deviation. Combine the individual deviation and the local consistency deviation to obtain the anomaly score of each point. Use the kernel density estimation method to determine the adaptive threshold of the anomaly score. Mark the points whose anomaly scores exceed the threshold as anomalies. Output the set of anomaly regions, including the location of the anomaly points and the corresponding anomaly scores.

[0091] In this embodiment, the construction of the k nearest neighbor graph and the calculation of the normalized Laplace matrix are important improvements. It can effectively capture the local structural features of the data, laying the foundation for subsequent manifold learning and semi-supervised classification. Especially when dealing with high-dimensional, nonlinear geological data, this graph-based representation method can better retain the essential structure of the data. The design of the objective function is another improvement. By comprehensively considering the fitting terms, graph regularization terms and function smoothing terms of the labeled data, the knowledge transfer from limited labeled samples to a large number of unlabeled samples is achieved. This semi-supervised learning strategy not only reduces the workload of manual annotation, but also effectively utilizes the structural information in the unlabeled data to improve the accuracy and generalization ability of classification. The design of the anomaly detection algorithm reflects innovation from multiple angles. By combining individual deviations and local consistency deviations, the abnormality of the data point itself and the incoordination with the surrounding environment can be considered at the same time. The application of the kernel density estimation method realizes the adaptive determination of the anomaly threshold, overcoming the limitations of the traditional fixed threshold method. High-precision geological structure classification and anomaly detection provide a comprehensive and reliable basis for assessing construction risks. For example, in subway tunnel construction, the ability to accurately identify potential risk points such as weak interlayers and karst development areas can help develop targeted reinforcement plans. For large underground projects, such as underground shopping malls or parking lot construction, this fine classification and anomaly detection capability can help engineers accurately assess the geological conditions in different areas and optimize spatial layout and support design. In complex geological environments, such as fault fracture zones or karst areas, it can effectively identify potential geological disaster points such as caves and underground rivers, providing key basis for risk assessment and preventive measures. In addition, in long-term construction monitoring, classification results and anomaly detection can be continuously updated to reflect changes in geological conditions in real time, which helps to promptly discover and respond to new problems that may arise during construction. In general, this step enhances the comprehensiveness and reliability of construction impact assessment by providing high-quality geological structure classification and anomaly detection results. It enables engineers to more accurately identify and quantify various types of geological risks, thereby developing more scientific and targeted construction plans and risk management strategies, and improving the safety and efficiency of underground projects.

[0092] According to one aspect of the present application, in step S44, a three-dimensional impact index field is obtained; and a three-dimensional isosurface generation algorithm module including an adaptive grid refinement module is used to generate isosurfaces of different impact levels, specifically:

[0093] Acquire 3D impact indicator field data. Perform adaptive spatial decomposition to improve accuracy in critical areas. Reconstruct implicit surfaces to represent isosurfaces. Extract isosurfaces on multi-resolution structures to generate triangular mesh models. Perform post-processing and visualization to output the final isosurface model.

[0094] Step S441, obtain three-dimensional impact indicator field data. Initialize a uniform cubic grid. Calculate the indicator field gradient in each grid cell. Set a gradient threshold and subdivide cells with gradients greater than the threshold. Recursively perform the subdivision process until the preset maximum depth or accuracy requirement is reached. Maintain the octree structure during the subdivision process and record the position and size information of each cell. Output the adaptively refined octree structure.

[0095] Step S442, obtain an adaptive octree structure. Sample points on the octree leaf nodes and calculate the signed distance function value. Select an appropriate radial basis function as the interpolation kernel. Construct a system of linear equations and solve the weight coefficients of the radial basis function. Add low-order polynomial terms to improve the stability of the interpolation. Use the solved coefficients to construct the implicit surface function of the entire space. Output the reconstructed implicit surface function.

[0096] Step S443, obtain implicit surface function and adaptive octree structure. For each leaf node of the octree, calculate the function value of the vertex. Determine whether the isosurface intersects with the current unit. If intersecting, calculate the intersection of the isosurface and the unit edge. Based on the intersection information, generate triangle fragments representing the isosurface. For adjacent units with different resolutions, ensure that the generated triangle fragments are seamlessly connected. Use the edge collapse algorithm to simplify the mesh while maintaining important features. Output a multi-resolution triangular mesh model.

[0097] Step S444, obtain a multi-resolution triangular mesh model. Apply the Laplace smoothing algorithm to improve the mesh quality. Calculate the normal vector of each vertex and use the weighted average method to ensure smooth transition. Estimate the average curvature and Gaussian curvature of each vertex. Assign a color value to each vertex according to a predefined influence level function. Construct vertex buffer objects and index buffer objects for efficient rendering. Generate normal maps and curvature maps to enhance visual effects. Output the post-processed color-coded multi-resolution isosurface model, including geometric information, topological information, normal vectors, curvature, and color mapping.

[0098] In this embodiment, the application of adaptive spatial decomposition technology is an important improvement. By dynamically adjusting the octree structure, the computational complexity can be effectively controlled while maintaining high precision in key areas. This adaptive strategy is particularly suitable for dealing with the non-uniform distribution characteristics of construction impacts, such as providing higher resolution around building foundations or near geological weak surfaces. The use of implicit surface reconstruction methods is another improvement. By combining radial basis function interpolation and low-order polynomial terms, smooth and continuous three-dimensional isosurfaces can be generated based on limited sampling points. Not only the quality of the isosurface is improved, but also the reliability of the model in data sparse areas is enhanced. The generation and post-processing technology of multi-resolution triangular mesh models embodies comprehensive innovative thinking. By ensuring seamless connection between units of different resolutions and applying techniques such as Laplace smoothing and curvature estimation, a high-quality, visually friendly three-dimensional model is achieved. The introduction of color coding further enhances the information transmission capability of the model, making the degree of impact clear at a glance. High-precision, intuitive three-dimensional visualization models provide powerful decision support tools for complex construction impact assessments. For example, in urban subway construction, this model can clearly show the spatial distribution and intensity changes of ground subsidence, helping decision makers to quickly identify high-risk areas and formulate targeted protection measures. For large-scale underground water conservancy projects, such as underground reservoir construction, the model can intuitively display the scope and intensity of the project's impact on the groundwater flow field, which helps to assess the potential impact on the surrounding ecological environment and surface buildings. In underground projects in historical and cultural heritage protection areas, this high-precision visualization model can accurately show the extent of the impact of construction activities on cultural relics protection units, providing a scientific basis for the formulation of protection strategies. In addition, in long-term construction monitoring, this visualization technology can dynamically display the spatiotemporal evolution of the impact range, helping engineers to adjust construction strategies in a timely manner and minimize adverse effects. In general, this step significantly improves the intuitiveness and operability of construction impact assessment by providing high-quality, information-rich 3D visualization models. It provides decision makers with a powerful tool to quickly and comprehensively understand the complex underground environment and construction impacts, so as to make more scientific and reasonable decisions and effectively balance the needs of engineering construction and environmental protection.

[0099] In some embodiments, the formula or model used is as follows:

[0100] The adaptive window size W = round(λ / Δx), where λ is the wavelength of the corresponding frequency and Δx is the sampling interval.

[0101] Adaptive gain function G(t) = a * exp(b * t c ) for gain correction. Where t is the two-way travel time, a, b, c are adaptive parameters determined by minimizing the signal energy variance.

[0102] Notch filter H(ω) = 1 - exp(-((ω - ω0) 2 / (2σ 2 ))) Where ω0 is the interference frequency and σ is the adjustable bandwidth parameter.

[0103] Weight function α(f, d) = (1 / f) * exp(-d / d0); where f is the frequency, d is the depth, and d0 is the feature depth parameter.

[0104] Velocity function v(x, y, z) = v0 + az + bexp(-c*z), where v0 is the surface velocity and a, b, c are fitting parameters.

[0105] The fusion function F(v, b) = w1v + w2b + w3*f(v, b), where v and b are GPR and BIM data respectively, f(v, b) is the interaction term, and w1, w2, and w3 are adaptive weights.

[0106] Anomaly score S(x) = (1 / k) * Σ_k_nearest d(x, xi) 2 , where d is the Mahalanobis distance.

[0107] Structural element SE(t) = {-aexp(-t 2 / 2σ 2 ), ..., 0, ..., aexp(-t 2 / 2σ 2 )} where a and σ are adaptively adjusted with the signal frequency.

[0108] Velocity field objective function: J(v) = Σi ||τi - τobsi|| 2 + λ * R(v) where τi is the calculated travel time, τobsi is the observed travel time, R(v) is the regularization term, and λ is the coefficient.

[0109] Anomaly score: S(x_i) = ||f*(x_i) - y_i|| 2 + α * Σ_j w_ij * ||f*(x_i) - f*(x_j)|| 2 , where α is the balance parameter. x_i represents the i-th data point. f*(x_i) represents the output value of the optimized classification function f* at the data point x_i. y_i means that if x_i is a labeled data point, y_i represents its true label; if x_i is an unlabeled data point, y_i can be understood as its predicted label. w_ij represents the edge weight between data points x_i and x_j, which is usually calculated based on their distance in the feature space.

[0110] ||f*(x_i) - y_i|| 2 : Represents the squared difference between the output value of the data point x_i and its label (or predicted label), measuring individual deviation.

[0111] According to another aspect of the present application, a construction impact identification system using geological radar technology is also provided, which is characterized by comprising:

[0112] at least one processor; and,

[0113] a memory communicatively connected to at least one of the processors; wherein,

[0114] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the construction impact identification method using geological radar technology as described in any of the above technical solutions.

[0115] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. The construction impact identification method using geological radar technology is characterized by: The steps include: Step S1, obtaining the original data set of each frequency collected by the multi-frequency geological radar system, and performing preprocessing including time zero point correction, direct wave elimination and gain correction; for the preprocessed data set, frequency domain adaptive filtering and space-time adaptive filtering are performed to eliminate interference from complex environments; multi-frequency data fusion is performed based on the filtered data set to obtain the final fused data; wherein the original data set of each frequency includes the original data of low frequency, medium frequency and high frequency; Step S2, obtaining the final fusion data and performing time-depth conversion to generate depth domain data; using the interpolation algorithm to perform 3D volume reconstruction, obtaining 3D GPR image data and importing it into the BIM model data of the project, and spatially registering and aligning the BIM model data with the 3D GPR image data; Call the data fusion algorithm module to perform semantic-level fusion of the aligned GPR data and BIM data to generate a fused 3D model; Step S3, obtaining the fused 3D model data and extracting a multi-scale feature set from it; using an unsupervised clustering algorithm to preliminarily classify the multi-scale feature set; calling a pre-configured semi-supervised learning algorithm module, and performing fine classification in combination with the labeled data; calling an anomaly detection algorithm module, and identifying potential abnormal areas based on the fine classification results; Step S4, obtaining the classification results and abnormal area information in step S3; extracting multi-level features from micro to macro scales; Call the cross-scale feature fusion network to integrate information at different scales; construct a spatiotemporal evolution model for the integrated multi-scale information, predict geological changes during the construction process, and obtain evolution results; design construction impact assessment indicators based on the evolution results; generate a three-dimensional impact gradient grid, visualize and output the impact range of different construction intensities; Step S2 is specifically as follows: Step S21, obtaining the final multi-frequency fusion data; calling the adaptive speed function considering the spatial position factor to perform time-depth conversion on the fusion data to generate depth domain data; using the improved interpolation algorithm module considering the spatial correlation to perform 3D volume reconstruction on the depth domain data to generate high-resolution 3D GPR image data; Step S22, obtaining the BIM model data of the project; and importing the 3D GPR image data; calling the spatial registration algorithm module including the feature point extraction and matching module to accurately align the 3D GPR image data and the BIM model data; Generate aligned GPR dataset and BIM dataset to maintain the integrity and spatial consistency of the original data; Step S23, obtaining the aligned GPR data set and BIM data set; calling a data fusion algorithm module including feature extraction, matching and fusion modules to extract geological features from the GPR data and extract geometric and semantic features from the BIM data; Calling a fusion function that considers the weights and interactions of data from different sources to process the extracted features and generate a fused 3D model that includes geological and architectural information; Step S4 is specifically as follows: Step S41, obtaining the final classification results and abnormal area information; extracting features from three scales, micro, meso and macro, including texture, structure and geological unit features; inputting the extracted features into the cross-scale feature fusion network of the attention mechanism, realizing feature fusion through adaptive weights; outputting the fused multi-scale features; Step S42, obtaining the fused multi-scale features; using a numerical method to solve the spatiotemporal evolution equation including the diffusion term, the reaction term and the attenuation term to obtain the evolution state at different time points; interpolating and smoothing the evolution results to generate continuous spatiotemporal evolution data; Step S43, obtaining spatiotemporal evolution data; calculating the impact index value of each spatial point at different times according to the preset construction impact assessment index; mapping the index value to different impact levels according to the set impact level classification standard; generating a three-dimensional impact index field; wherein the construction impact assessment index includes a state change index, a spatial gradient index and a curvature index; and the three-dimensional impact index field includes spatial coordinates and corresponding impact index values; Step S44, obtaining a three-dimensional impact index field; using a three-dimensional isosurface generation algorithm module including an adaptive mesh refinement module to generate isosurfaces of different impact levels; constructing a three-dimensional impact gradient mesh; superimposing the generated isosurfaces and gradient meshes with the BIM model data; and outputting a final three-dimensional construction impact visualization model; Among them, each grid cell in the three-dimensional impact gradient grid contains location coordinates and impact level information; the three-dimensional construction impact visualization model contains underground structure, impact isosurface and color-coded impact gradient grid.

2. The construction impact assessment method using geological radar technology as claimed in claim 1, characterized in that: The step S1 is specifically as follows: Step S11, using a multi-frequency geological radar system to simultaneously collect data of each frequency, including low-frequency, medium-frequency and high-frequency data; on a preset survey line, using antennas of different frequencies to collect data; setting a sampling interval and time window for each frequency; recording the two-way travel time and signal amplitude of each sampling point; storing the collected low-frequency, medium-frequency and high-frequency data as original data sets, including the number of sampling points, sampling interval and time window; Step S12, obtaining the original data set; performing time zero point correction on the data set of each frequency to ensure that the signal start time is consistent; Calculate the wavelength and sampling interval corresponding to each frequency and determine the adaptive window size; use the median filter method with adaptive window size to process the low-frequency, medium-frequency and high-frequency data to remove the direct wave; call the constructed adaptive gain function with adjustable parameters and apply it to the data after removing the direct wave to obtain the gain-corrected data set; Step S13, obtaining a data set after gain correction; performing a fast Fourier transform on the data set of each frequency to obtain frequency domain data; calling an adaptive notch filter including adjustable interference frequency and bandwidth parameters, applying it to the frequency domain data, and performing a filtering operation; Performing inverse fast Fourier transform on the filtered frequency domain data to obtain time domain data; calling a four-dimensional filter including a spatial weight function and a time weight function and applying it to the time domain data to obtain each frequency data after removing the interference of complex environment; Step S14, obtaining each frequency data after removing the interference of complex environment; performing multi-scale wavelet decomposition on the data set of each frequency to obtain respective approximate coefficients and detail coefficients; calling the constructed adaptive weight function considering frequency and depth factors; and using the adaptive weight function to fuse the approximate coefficients and detail coefficients of different frequency data; using the fused approximate coefficients and detail coefficients to perform inverse wavelet transform to obtain the final multi-frequency fusion data.

3. The construction impact assessment method using geological radar technology as claimed in claim 2 is characterized in that: Step S31, obtaining a fused 3D model; using a pre-configured multi-scale feature extractor including spatial, frequency and texture features to process the 3D model data to generate a multi-dimensional feature vector; normalizing and reducing the dimensionality of the extracted multi-dimensional feature vector to obtain a final feature set; Step S32, obtaining a final feature set; using a density peak clustering algorithm module including an adaptive density estimation module and a relative density difference calculation module to perform cluster analysis on the final feature set to generate a preliminary classification result; performing a post-processing step to merge similar clusters and eliminate isolated points to obtain an optimized clustering result; Step S33, obtaining preliminary classification results and a small amount of manually labeled sample data; constructing a graph model, calculating the similarity between samples; calling a semi-supervised learning algorithm module including a label propagation and manifold regularization module to refine the preliminary classification results to obtain a final classification result; Based on the final classification results, calculate the anomaly score of each sample point; And identify potential abnormal areas according to the set adaptive threshold, and output the abnormal area set.

4. The construction impact assessment method using geological radar technology as claimed in claim 3 is characterized in that: In step S12, a median filter method with an adaptive window size is used to process the low-frequency, medium-frequency and high-frequency data to remove the direct wave, specifically: Step S121, obtaining low-frequency, medium-frequency and high-frequency data sets after time zero point correction; Step S122: construct an adaptive structural element according to the signal characteristics of each frequency, and the shape parameters are dynamically adjusted according to the signal frequency; Step S123, applying an opening operation to each frequency data set to remove the peaks in the data; applying a closing operation to the result of the opening operation to fill the valleys in the data; The results of the opening operation and the closing operation are added and averaged to obtain the estimated background signal; Step S124, calculating the difference between the original data and the estimated background signal to obtain a residual signal after removing the direct wave; Outputs processed low-frequency, mid-frequency, and high-frequency data sets with direct wave components removed.

5. The construction impact assessment method using geological radar technology as claimed in claim 3 is characterized in that: In step S13, an adaptive notch filter is used to perform filtering operations, specifically: Step S131, obtaining transformed low-frequency, medium-frequency and high-frequency data sets; performing wavelet packet transform on each data set; using a predetermined wavelet basis function and decomposition layer number, performing wavelet packet decomposition to obtain a time-frequency representation of each frequency data; outputting wavelet packet coefficients of low-frequency, medium-frequency and high-frequency data; Step S132, obtaining the wavelet packet coefficients of each frequency data set; calculating the noise standard deviation of each decomposition level and node; defining an adaptive threshold for each wavelet packet coefficient according to the signal length and the noise standard deviation; According to the designed soft threshold function, the wavelet packet coefficients are threshold processed; Output denoised wavelet packet coefficients; Step S133, obtain the denoised wavelet packet coefficients; select the same wavelet basis function as that used in the decomposition; perform inverse wavelet packet transform to reconstruct the time domain signal; obtain filtered data of low frequency, medium frequency and high frequency respectively; and output the filtered low frequency, medium frequency and high frequency data sets.

6. The construction impact assessment method using geological radar technology as claimed in claim 3 is characterized in that: In step S14, multi-scale wavelet decomposition is performed on the data set of each frequency to obtain respective approximate coefficients and detail coefficients, specifically: Step S141, obtaining each frequency data after removing the interference of the complex environment, performing empirical mode decomposition on the data set of each frequency; iteratively extracting each intrinsic mode function until the remaining signal becomes a monotonic function; recording the intrinsic mode function set of each frequency data; outputting the intrinsic mode function set of low-frequency, medium-frequency and high-frequency data; Step S142, obtaining the intrinsic mode function set of each frequency; selecting a predetermined wavelet function; performing continuous wavelet transform on each intrinsic mode function; calculating wavelet coefficients of different scales and time positions; outputting the continuous wavelet transform results of all intrinsic mode functions of each frequency data; Step S143, obtaining a continuous wavelet transform result; Calculate the energy distribution of each frequency signal at different scales; design an adaptive weight function based on the energy distribution; use the weight function to fuse the wavelet transform results of different frequencies; perform an inverse transform on the fused wavelet transform results to separate the approximate coefficients and detail coefficients; output the fused time-frequency representation, as well as the corresponding approximate coefficients and detail coefficients.

7. The construction impact assessment method using geological radar technology as claimed in claim 3 is characterized in that: In step S21, the final multi-frequency fusion data is obtained; the adaptive speed function considering the spatial position factor is called to perform time-depth conversion on the fusion data to generate depth domain data, specifically: Step S211, obtaining the final multi-frequency fusion data, using the dynamic time warping algorithm to calculate the similarity between local waveforms; constructing a similarity matrix; applying the spectral clustering algorithm to group the waveforms based on the similarity matrix; and outputting the waveform clustering results; Step S212, obtaining waveform clustering results; For each cluster, estimate the velocity using a similarity analysis method; calculate waveform corrections and similarity values ​​at different velocities; select the velocity with the largest similarity value as the estimated velocity for the cluster; construct an initial velocity field, including spatial positions and corresponding velocity estimates; Step S213, obtain the initial velocity field; apply the tensor voting framework to smooth and extrapolate the velocity field; design the objective function, including the travel time fitting term and the regularization term; use the L-BFGS algorithm to optimize the velocity field and minimize the objective function; obtain the optimized three-dimensional velocity field; use the optimized velocity field to perform time-depth conversion on the fused data of step S14; output the depth domain data.

8. A construction impact assessment system using geological radar technology, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the construction impact identification method using geological radar technology as described in any one of claims 1 to 7.

Images