Multi-dimensional display method and system based on intelligent analysis of three-dimensional geological radar data

Abstract

The present application belongs to the field of geoscience information technology, and relates to a multidimensional display method and system based on intelligent analysis of three-dimensional geological radar data. Firstly, the three-dimensional geological radar data is converted into images, and an accurate mapping relationship between image pixels and actual physical dimensions is established, realizing intelligent identification from image anomaly area interpretation to quantitative analysis. Secondly, based on a multidimensional collaborative visualization interface, the same view synchronously displays multi-channel vertical sections, multi-depth horizontal slices and synchronous video frames, integrates fragmented data dimensions through a time-space synchronous linkage mechanism, and then uses a target recognition model trained by multi-modal data to carry out fusion feature extraction analysis on the multi-channel vertical sections, multi-depth horizontal slices and synchronous video frames. The surface feature correlation degree is used as the key basis for determining the confidence degree, which can support or exclude underground anomalies, greatly reduce the false positive rate of traditional single image analysis method, and improve the radar recognition accuracy and reliability.

Description

Technical Field

[0001] This invention belongs to the field of geoscience information technology, and more specifically, relates to a multi-dimensional display method and system based on intelligent analysis of three-dimensional ground-penetrating radar data. Background Technology

[0002] As a non-destructive, rapid, and intuitive detection technology, ground-penetrating radar has shown broad application prospects and significant value in many fields such as urban underground space surveys, detection of underground defects and pipelines, analysis of road structural defects, and archaeological exploration.

[0003] Among them, the three-dimensional ground-penetrating radar, with its unique acquisition method, can simultaneously acquire vertical profile data from multiple channels and generate horizontal slice maps at different depths by fusing data from the same horizontal position, providing richer and more comprehensive underground information. It features multi-channel vertical profile maps and horizontal slice maps at various depths, greatly expanding the interpretation dimensions and reference basis of ground-penetrating radar data.

[0004] However, existing technologies have significant shortcomings in the processing and display of 3D geological radar data. Specifically, these shortcomings are as follows: 1. Although existing technologies can generate radar images, they generally fail to establish a high-precision mapping relationship between image pixels and actual physical dimensions. This results in the interpretation results only being at the level of image anomalies, and the true geometric dimensions of underground targets cannot be calculated directly and accurately. Consequently, data analysis is difficult to move from qualitative to quantitative analysis, and cannot meet the needs of accurate assessment and engineering decision-making.

[0005] 2. Existing software typically displays multi-channel vertical profiles, multi-depth horizontal slices, and synchronous video data in isolated windows or modules. There is no spatiotemporal correlation or interactive response between the views. Analysts need to manually correlate and compare a large amount of isolated data based on experience, which is extremely inefficient and makes it easy to miss key information. The advantages of 3D data in terms of multi-dimensionality and high information content are difficult to translate into advantages in interpretation efficiency.

[0006] 3. Most existing automated geological anomaly identification methods only analyze radar image signals, completely ignoring the surface environmental features recorded by synchronous video. They are difficult to distinguish between real signals generated by underground anomalies and false anomalies caused by surface interference, resulting in a high false alarm rate and making it difficult to guarantee the reliability and accuracy of the identification results. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background technology, a multi-dimensional display method and system based on intelligent analysis of three-dimensional ground-penetrating radar data is proposed.

[0008] The technical solution adopted by the present invention to solve its technical problem is as follows: Firstly, the present invention provides a multi-dimensional display method based on intelligent analysis of three-dimensional ground-penetrating radar data, including: collecting raw three-dimensional ground-penetrating radar data, synchronous video data and positioning data.

[0009] The original three-dimensional ground-penetrating radar data is preprocessed and combined with the positioning data to assign spatial coordinate information, thereby generating a ground-penetrating radar data sequence carrying spatial location attributes.

[0010] The synchronized video data is timestamped to generate a video frame sequence that is time-synchronized with the ground-penetrating radar data sequence.

[0011] Based on the aforementioned ground-penetrating radar data sequence, a multi-channel vertical profile dataset and a multi-depth horizontal slice dataset were constructed, and a mapping relationship between image pixels and actual physical dimensions was established.

[0012] A multi-dimensional collaborative visualization interface is constructed, which synchronously displays the multi-channel vertical profile, multi-depth horizontal slice, and synchronous video frames in the same view, and supports linkage response based on timestamp and spatial coordinates.

[0013] The multi-channel vertical profile, multi-depth horizontal slice, and synchronous video frames are intelligently fused and analyzed to identify geological anomalies and calculate their physical dimensions.

[0014] Output a detection report containing the physical size and spatial location information of the anomaly, and mark the three-dimensional location of the anomaly in the geospatial platform.

[0015] Secondly, the present invention provides a multi-dimensional display system based on intelligent analysis of three-dimensional ground-penetrating radar data, including: a data acquisition module, a radar data processing module, a video data alignment module, an image conversion module, a multi-dimensional display module, an intelligent recognition module, and a detection output module.

[0016] The data acquisition module is connected to the radar data processing module and the video data alignment module respectively. The radar data processing module is connected to the image conversion module. The image conversion module and the video data alignment module are both connected to the multi-dimensional display module. The multi-dimensional display module is connected to the intelligent recognition module. The intelligent recognition module is connected to the detection output module.

[0017] The data acquisition module collects raw 3D ground-penetrating radar data, synchronous video data, and positioning data.

[0018] The radar data processing module preprocesses the original three-dimensional ground-penetrating radar data, combines it with the positioning data to assign spatial coordinate information, and generates a ground-penetrating radar data sequence carrying spatial location attributes.

[0019] The video data alignment module performs timestamp alignment processing on the synchronized video data to generate a video frame sequence that is time-synchronized with the ground-penetrating radar data sequence.

[0020] The image conversion module, based on the ground-penetrating radar data sequence, constructs a multi-channel vertical profile dataset and a multi-depth horizontal slice dataset, and establishes a mapping relationship between image pixels and actual physical dimensions.

[0021] The multi-dimensional display module constructs a multi-dimensional collaborative visualization interface, which synchronously displays the multi-channel vertical cross-sectional view, multi-depth horizontal slice view, and synchronous video frames in the same view, and supports linkage response based on timestamps and spatial coordinates.

[0022] The intelligent recognition module performs intelligent fusion analysis on the multi-channel vertical profile, multi-depth horizontal slice, and synchronous video frames to identify geological anomalies and calculate their physical dimensions.

[0023] The detection output module outputs a detection report containing the physical size and spatial location information of the anomaly, and marks the three-dimensional location of the anomaly in the geospatial platform.

[0024] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention transforms three-dimensional ground-penetrating radar data into images and establishes a precise mapping relationship between image pixels and actual physical dimensions in the length, width and depth directions, so that the intelligent recognition results are no longer limited to abnormal areas at the image level, realizing the leap from image interpretation to quantitative analysis, and effectively improving the accuracy and engineering value of data analysis.

[0025] (2) By constructing a multi-dimensional collaborative visualization interface, the present invention can simultaneously display multi-channel vertical profiles, multi-depth horizontal slices and synchronous video frames in the same view, and establish a spatiotemporal synchronous linkage mechanism between them, integrating the originally fragmented data dimensions into an organic whole. Interpreters do not need to manually switch and compare between multiple windows. Operations on any dimension can trigger global linkage updates, making integrated and intuitive analysis of complex underground spaces possible.

[0026] (3) This invention uses a target recognition model trained with multimodal data to perform fusion feature extraction and analysis on radar images and synchronous video frames, and takes the correlation with surface features as the key basis for determining confidence. It effectively uses surface context information to corroborate or exclude underground anomalies, greatly overcoming the drawback of high false alarm rate of traditional single image analysis method, making the recognition results more accurate and reliable. Attached Figure Description

[0027] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the method flow provided in the first embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of radar data zero-point correction according to the first embodiment of the present invention.

[0030] Figure 3 This is a system module connection diagram provided for the second embodiment of the present invention. Detailed Implementation

[0031] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example 1

[0033] like Figure 1 As shown, the first embodiment of the present invention provides a multi-dimensional display method based on intelligent analysis of three-dimensional ground-penetrating radar data, including: S01. Acquiring raw three-dimensional ground-penetrating radar data, synchronous video data and positioning data.

[0034] It should be noted that the original three-dimensional ground-penetrating radar data is a multi-dimensional data set parsed from ground-penetrating radar data files. Its dimensional information includes at least the number of channels, the number of sampling points per radar data channel, the acquisition timestamp, the amplitude sequence, the survey line number, the sampling frequency, and the channel spacing. The number of channels is used to characterize the total amount of data collected along the radar survey line direction. The number of sampling points per radar data channel is used to characterize the total amount of data sampled by the radar in the depth direction. The amplitude sequence records the original electromagnetic response of the radar signal propagating and reflecting in the medium. The survey line number is used to identify the detection line to which the data belongs. The sampling frequency determines the resolution on the time axis, and its reciprocal corresponds to the size of the time window. The time window defines the maximum time interval from radar signal transmission to reception to determine the maximum detection depth. The channel spacing is the physical distance between adjacent detection channels.

[0035] The synchronized video data specifically refers to the surface video images continuously recorded in the area detected by the ground-penetrating radar. The camera ensures that each frame of video image can be accurately associated with the corresponding radar scanning position through a timestamp synchronization mechanism with the main control unit of the ground-penetrating radar. The frame rate of the video image should match or be an integer multiple of the scanning frequency of the ground-penetrating radar to ensure synchronization accuracy.

[0036] The positioning data is GPS coordinate data or location data generated based on an encoder.

[0037] S02. The original three-dimensional ground-penetrating radar data is preprocessed, and spatial coordinate information is assigned in combination with the positioning data to generate a ground-penetrating radar data sequence carrying spatial location attributes.

[0038] In a preferred embodiment of the present invention, the preprocessing of the original three-dimensional ground-penetrating radar data includes: parsing the data format type of the original three-dimensional ground-penetrating radar data, and loading the corresponding preprocessing process according to the parsed data format type. The preprocessing process includes at least background noise removal, gain processing, filtering processing, and data zero-point correction.

[0039] It should be noted that the above data format type parsing relies on a configured data file interface module, which communicates with the data storage medium through a preset interface protocol. When a raw 3D ground-penetrating radar (GPR) data file is received, the file content parsing process is initiated. This process first reads a specific byte sequence and field structure in the file header area to identify the file's magic number, version identifier, data block layout, and internal encoding scheme. It then retrieves the built-in data format registry in the web cloud, which includes, but is not limited to, the characteristic signatures of various common GPR data formats such as GSSI DZT format, Mala*.rad format, Sensors & Software DZC / DZT format, and data formats conforming to the SEG-Y standard. The data file interface module compares the parsed file header information with the data format registry to automatically determine the specific format type of the current GPR data file.

[0040] The aforementioned background noise removal eliminates constant background noise unrelated to underground targets by using moving average or linear fitting methods.

[0041] Gain processing compensates for the echo signal intensity using exponential gain or time gain methods, based on the attenuation characteristics of electromagnetic waves in underground media, to enhance the visibility of deep targets.

[0042] Filtering uses digital filters, such as Butterworth filters or Gaussian filters, to remove high-frequency random noise and low-frequency system drift from radar signals, while retaining the signal within the effective frequency band.

[0043] like Figure 2 As shown, the data zero-point correction involves retrieving specific signal characteristics of each radar data channel within a preset start time window to determine the actual signal zero point of each radar data channel. The sampling point data before the actual signal zero point is then cut off or zeroed out, and the remaining data is time-shifted and resampled to align the signal zero points of all radar data channels in the time dimension, generating a radar data volume with consistent depth direction.

[0044] It should be noted that the specific signal features retrieved by the above data zero-point correction specifically refer to the search for the first occurrence of a high amplitude or high gradient signal within a preset time window after the radar signal is emitted from the antenna. This signal usually corresponds to the propagation of the radar direct wave between antennas or the reflection when the radar pulse first contacts the surface of the medium. High amplitude or high gradient can be exemplarily referred to as exceeding three times the standard deviation of the mean of the corresponding parameter within the preset time window.

[0045] It should be added that the reason for zero-point correction is that in the actual operation of ground-penetrating radar, due to factors such as surface undulation, noise interference in the detection environment, and the coupling effect of the radar equipment itself, it is impossible to ensure that the zero-point position in the original data waveform of each survey line is completely consistent with the theoretical medium surface position. The zero-point position is the starting time point when the radar signal is emitted from the antenna and begins to propagate. In order to eliminate this kind of systematic error in the depth direction and ensure the alignment and consistency of the data of each survey line in the depth dimension.

[0046] In a preferred embodiment of the present invention, the process of generating the ground-penetrating radar data sequence carrying spatial location attributes includes: parsing the positioning data and obtaining the timestamp and latitude and longitude coordinates of each positioning point.

[0047] Extract the acquisition timestamp of each radar data channel from the original three-dimensional ground-penetrating radar data.

[0048] Each radar data channel is matched with multiple neighboring positioning points based on timestamps, and the spatial coordinates of each radar data channel are calculated using an interpolation algorithm.

[0049] It should be noted that the specific process for obtaining the spatial coordinates of each radar data channel is as follows: For radar data channels whose acquisition timestamps are located within the time interval between two consecutive positioning points, a linear interpolation algorithm is used to calculate the precise spatial coordinates of the data channel on the line connecting the two consecutive positioning points based on the relative position of the acquisition timestamp of the data channel within the time interval.

[0050] The generation of ground-penetrating radar (GPR) data sequences carrying spatial location attributes is a crucial step in ensuring the accuracy of geological exploration results. Its core significance stems from the issue of acquisition frequency differences caused by the physical characteristics of the equipment: In actual exploration scenarios, the data acquisition frequency of positioning equipment is limited by physical conditions such as hardware response speed and signal reception intervals, resulting in a data acquisition frequency far lower than that of the GPR host. Typically, within the time interval between the positioning equipment recording one spatial coordinate point, the GPR has already completed the acquisition of hundreds or even thousands of radar data channels. This frequency difference directly leads to the vast majority of radar data channels lacking precise time-synchronized positioning information. If a simple nearest-neighbor assignment method is used—that is, directly matching the unsynchronized radar data channels to the nearest positioning point—significant sawtooth errors will occur in the spatial location of the radar profile. This error is not a random disturbance but exhibits periodic positional jumps, which not only distorts the actual spatial distribution of underground targets but may also obscure the true location information of small geological anomalies, ultimately severely affecting the accuracy of geological interpretation and even leading to biased exploration conclusions.

[0051] The calculated spatial coordinates are used as attribute information and associated with the corresponding radar data channels to form a geological radar data sequence carrying spatial location attributes.

[0052] S03. Perform timestamp alignment processing on the synchronized video data to generate a video frame sequence that is time-synchronized with the ground-penetrating radar data sequence.

[0053] S04. Based on the aforementioned ground-penetrating radar data sequence, construct a multi-channel vertical profile dataset and a multi-depth horizontal slice dataset, and establish a mapping relationship between image pixels and actual physical dimensions.

[0054] In a preferred embodiment of the present invention, the process of constructing the multi-channel vertical profile dataset includes: taking the radar data of each detection channel as input, determining the image width based on the number of channels, determining the image height based on the number of sampling points, and constructing a pixel grid of a two-dimensional image.

[0055] For each radar data channel, a single-column pixel mapping is performed. Based on the acquisition order of the amplitude values ​​of the data channel, the amplitude value sequence is matched one by one to the vertical axis coordinate position of the corresponding image column pixel.

[0056] By using a predetermined color mapping mechanism, the amplitude value of each sampling point is independently converted into the corresponding RGB color value and filled into its mapped pixel position, thereby generating a vertical profile spectrum image of each channel. The color mapping mechanism includes normalizing the amplitude value and applying a pseudo-color or grayscale mapping table.

[0057] In a preferred embodiment of the present invention, the process of constructing the multi-depth horizontal slice map dataset includes: determining the actual depth corresponding to each sampling point based on radar wave velocity and sampling frequency parameters.

[0058] It should be noted that the detection principle of ground-penetrating radar is to transmit high-frequency electromagnetic pulses into the ground through a transmitting antenna. When the wave encounters the interface between different media, it will be reflected, and then the receiving antenna will receive the reflected wave. The above-mentioned actual depth calculation process corresponds to this physical process: the ratio of the sampling point index to the sampling frequency is used as the propagation time of the reflected wave, the product of the radar wave speed and the propagation time of the reflected wave is used as the two-way path of the radar wave, and the value of the one-way path of the radar wave is used as the actual depth.

[0059] For each target depth, extract the pixel row corresponding to that depth from the vertical profile atlas image of all channels.

[0060] The extracted pixels in each row are stitched together according to their corresponding channel's physical spatial location to generate a horizontal slice image representing the structural distribution of that depth level.

[0061] In a preferred embodiment of the present invention, establishing the mapping relationship between image pixels and actual physical size includes: determining the physical length value corresponding to the pixels of the image in the detection direction based on the channel spacing.

[0062] The physical depth value corresponding to a pixel in the depth direction of the image is determined based on the sampling frequency, radar wave velocity, and number of sampling points.

[0063] It should be noted that the physical depth value corresponding to the pixel in the depth direction of the above image specifically refers to the result of the ratio of the actual depth quantized by the sampling frequency and radar wave speed to the number of sampling points.

[0064] The physical width value of each pixel in the width direction of the image is determined based on the number of channels and the channel spacing.

[0065] It should be noted that the physical width value of the pixel in the width direction of the above image is specifically the result of multiplying the difference between the number of channels and 1 by the channel spacing.

[0066] Based on the determined physical size mapping relationship, corresponding physical scales are generated and added to the vertical profile atlas image and the horizontal slice atlas image.

[0067] This invention converts 3D ground-penetrating radar data into images and establishes a precise mapping relationship between image pixels and actual physical dimensions in the length, width, and depth directions. This enables intelligent identification results to move beyond abnormal areas at the image level, achieving a leap from image interpretation to quantitative analysis and effectively improving the accuracy and engineering value of data analysis.

[0068] S05. Construct a multi-dimensional collaborative visualization interface, which synchronously displays the multi-channel vertical profile, multi-depth horizontal slice, and synchronous video frames in the same view, and supports linkage response based on timestamps and spatial coordinates.

[0069] In a preferred embodiment of the present invention, the construction of the multi-dimensional collaborative visualization interface includes:

[0070] Set up a vertical profile display area, a horizontal slice display area, and a video playback area to display the selected vertical profile, a horizontal slice at a specific depth, and a synchronized video frame, respectively.

[0071] Establish a spatiotemporal linkage mechanism among the three regions, and respond to user interaction operations in any region to synchronously update the displayed content in the other regions.

[0072] The linkage mechanism specifically achieves the following synchronization: i. Based on the vertical position selected on the vertical profile, the horizontal slice display area is synchronously updated to the slice at the corresponding depth, and the video playback area is controlled to jump to the video frame at the corresponding time.

[0073] ii. Based on the horizontal position selected on the horizontal slice map, synchronously switch the vertical profile display area to the vertical profile passing through that position and control the video playback area to jump to the video frame at the corresponding time.

[0074] iii. Based on the adjustment of the video playback progress, synchronously update the vertical profile and horizontal slice display areas to the radar image corresponding to the current video frame time and position.

[0075] It should be noted that the aforementioned multi-dimensional collaborative visualization interface also supports data linkage and viewing functions. Users can input a specific track number or survey line travel distance, and the interface will automatically position all views to that location for display.

[0076] This invention constructs a multi-dimensional collaborative visualization interface that synchronously displays multi-channel vertical profiles, multi-depth horizontal slices, and synchronous video frames in the same view, and establishes a spatiotemporal synchronization linkage mechanism between them. This integrates the originally fragmented data dimensions into an organic whole, eliminating the need for interpreters to manually switch and compare between multiple windows. Operations on any dimension can trigger global linkage updates, making integrated and intuitive analysis of complex underground spaces possible.

[0077] S06. Perform intelligent fusion analysis on the multi-channel vertical profile, multi-depth horizontal slice, and synchronous video frames to identify geological anomalies and calculate their physical dimensions.

[0078] In a preferred embodiment of the present invention, the geological anomaly identification process includes: training a multimodal deep learning network based on a geological radar sample dataset labeled with anomaly type, boundary information and corresponding surface environment features to obtain a target identification model.

[0079] The multi-channel vertical profile, multi-depth horizontal slice, and synchronous video frames are input into the target recognition model. The model performs fusion feature extraction and analysis to identify potential abnormal targets in the image and outputs the type and confidence level corresponding to each recognition result.

[0080] It should be noted that the geological anomaly identification process utilizes the feature extraction network of the target identification model to extract depth features representing the reflection structure of the underground medium from multi-channel vertical profiles and multi-depth horizontal slices, and to extract visual features representing the surface environment and scene from synchronized video frames.

[0081] The model's built-in feature fusion unit performs cross-modal association and fusion of extracted radar depth features and visual features to generate a unified feature identifier that integrates underground information and surface context.

[0082] Based on the unified feature identifier, the model's classification and region proposal network performs decoding: multiple candidate boxes of different sizes and proportions are generated at various locations in the image, and a subset of the unified feature identifier corresponding to each candidate box is extracted. The probability of the subset of the unified feature identifier within the candidate box belonging to each anomaly type is obtained. If the maximum probability value exceeds the preset category probability threshold, the category corresponding to the maximum probability value of the candidate box is initially determined and recorded as the target category.

[0083] The gradient of the target category relative to the radar depth features within the candidate box is calculated. The gradient information is then integrated and back-mapped to the original radar image. The contrast deviation of the candidate box relative to the background is quantified, and this is used as the salience of the abnormal target in the radar image.

[0084] For candidate bounding boxes identified as target categories, the cosine similarity between their visual feature vectors and typical samples in the predefined positive sample feature vector library of the target category is calculated. This similarity is used as the correlation between the abnormal target and the surface features in the video frame.

[0085] The confidence level is determined based on the output probability of the classification layer in the target recognition model, the salience of the abnormal target in the radar image, and its correlation with the surface features in the video frame.

[0086] It should be noted that the confidence level mentioned above can be calculated through linear weighted fusion, where the allocation of linear weights is specifically the optimal weights learned by the target recognition model through a series of correct decision samples.

[0087] Identification results that reach a preset confidence threshold are judged as valid anomalies.

[0088] In a preferred embodiment of the present invention, the physical size calculation process includes: delineating the boundaries of the identified abnormal body based on the boundary information of the abnormal body output by the target recognition model.

[0089] The number of pixels occupied by the boundary delineation area along the length, width, and depth directions in the vertical cross-sectional view and the horizontal slice view are counted respectively.

[0090] By combining the pre-established mapping relationship between image pixels and actual physical size, the number of pixels is multiplied by the corresponding physical value per unit pixel to calculate the actual length, actual width, and actual depth of the anomaly.

[0091] This invention employs a target recognition model trained with multimodal data to perform fusion feature extraction and analysis on radar images and synchronized video frames. It uses the correlation with surface features as the key basis for determining confidence, effectively utilizing surface context information to corroborate or exclude underground anomalies. This greatly overcomes the drawback of high false alarm rate in traditional single image analysis methods, making the recognition results more accurate and reliable.

[0092] S07. Output a detection report containing the physical size and spatial location information of the anomaly, and mark the three-dimensional location of the anomaly in the geospatial platform.

[0093] Example 2

[0094] like Figure 3 As shown, the second embodiment of the present invention provides a multi-dimensional display system based on intelligent analysis of three-dimensional ground-penetrating radar data, including: a data acquisition module, a radar data processing module, a video data alignment module, an image conversion module, a multi-dimensional display module, an intelligent recognition module, and a detection output module.

[0095] The data acquisition module is connected to the radar data processing module and the video data alignment module respectively. The radar data processing module is connected to the image conversion module. The image conversion module and the video data alignment module are both connected to the multi-dimensional display module. The multi-dimensional display module is connected to the intelligent recognition module. The intelligent recognition module is connected to the detection output module.

[0096] The data acquisition module collects raw 3D ground-penetrating radar data, synchronous video data, and positioning data.

[0097] The radar data processing module preprocesses the original three-dimensional ground-penetrating radar data, combines it with the positioning data to assign spatial coordinate information, and generates a ground-penetrating radar data sequence carrying spatial location attributes.

[0098] The video data alignment module performs timestamp alignment processing on the synchronized video data to generate a video frame sequence that is time-synchronized with the ground-penetrating radar data sequence.

[0099] The image conversion module, based on the ground-penetrating radar data sequence, constructs a multi-channel vertical profile dataset and a multi-depth horizontal slice dataset, and establishes a mapping relationship between image pixels and actual physical dimensions.

[0100] The multi-dimensional display module constructs a multi-dimensional collaborative visualization interface, which synchronously displays the multi-channel vertical cross-sectional view, multi-depth horizontal slice view, and synchronous video frames in the same view, and supports linkage response based on timestamps and spatial coordinates.

[0101] The intelligent recognition module performs intelligent fusion analysis on the multi-channel vertical profile, multi-depth horizontal slice, and synchronous video frames to identify geological anomalies and calculate their physical dimensions.

[0102] The detection output module outputs a detection report containing the physical size and spatial location information of the anomaly, and marks the three-dimensional location of the anomaly in the geospatial platform.

[0103] The multi-dimensional display system based on intelligent analysis of three-dimensional ground-penetrating radar data provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0104] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A multi-dimensional display method based on intelligent analysis of three-dimensional ground penetrating radar data, characterized in that, The method comprises the following steps: Collecting original three-dimensional ground penetrating radar data, synchronous video data and positioning data; Preprocessing the original three-dimensional ground penetrating radar data, combining the positioning data to give spatial coordinate information, and generating a ground penetrating radar data sequence carrying spatial position attributes; Timestamp alignment processing of the synchronous video data is performed to generate a video frame sequence that is time-synchronized with the ground penetrating radar data sequence; Based on the ground penetrating radar data sequence, a multi-channel vertical profile data set and a multi-depth horizontal slice data set are constructed, and a mapping relationship between image pixels and actual physical dimensions is established; A multi-dimensional collaborative visualization interface is constructed, which synchronously displays the multi-channel vertical profile, multi-depth horizontal slice and synchronous video frame in the same view, and supports timestamp and spatial coordinate based linkage response; Intelligent fusion analysis is performed on the multi-channel vertical profile, multi-depth horizontal slice and synchronous video frame to identify geological anomaly bodies and calculate their physical dimensions; The geological anomaly body identification process comprises: Based on a ground penetrating radar sample data set labeled with anomaly body types, boundary information and corresponding surface environment characteristics, a multi-modal deep learning network is trained to obtain a target recognition model; The multi-channel vertical profile, multi-depth horizontal slice and synchronous video frame are input into the target recognition model, which extracts and analyzes the fusion features to identify potential anomaly targets in the image, and outputs the type and confidence of each identification result; In the geological anomaly identification process, the feature extraction network of the target recognition model is used to extract depth features representing underground medium reflection structures from the multi-channel vertical profile and multi-depth horizontal slice, and visual features representing the surface environment and scene from the synchronous video frame; Through the feature fusion unit built-in the model, the extracted radar depth features and visual features are cross-modally associated and fused to generate a unified feature identifier that integrates underground information and surface context; Based on the unified feature identifier, the classification and region proposal network of the model decodes: multiple candidate boxes of different sizes and proportions are generated at each position of the image, and the unified feature identifier subset corresponding to each candidate box is extracted to obtain the probability that the unified feature identifier subset belongs to each anomaly body type, and if the maximum probability value exceeds the pre-set class probability threshold, the class corresponding to the maximum probability value is preliminarily determined as the target class; The gradient of the target class relative to the radar depth features in the candidate box is calculated, the gradient information is reversely mapped to the original radar image, and the contrast deviation of the candidate box relative to the background is quantified as the saliency of the anomaly target in the radar image; For the candidate box identified as the target class, the cosine similarity between its visual feature vector and the typical sample feature vector in the target class pre-defined positive sample feature vector library is calculated as the association degree of the anomaly target and the surface feature in the video frame; The confidence is determined based on the output probability of the classification layer in the target recognition model, the saliency of the anomaly target in the radar image, and the association degree of the anomaly target and the surface feature in the video frame. The recognition result with the confidence reaching the preset threshold is determined as an effective anomaly body; A detection report containing physical size and spatial position information of the anomaly body is output, and a three-dimensional position of the anomaly body is marked in a geographic space platform. 2.The multi-dimensional presentation method based on intelligent analysis of three-dimensional ground penetrating radar data according to claim 1, characterized in that, The original three-dimensional geological radar data is preprocessed, including: The original three-dimensional geological radar data is analyzed in terms of data format type, a corresponding preprocessing procedure is loaded according to the analyzed data format type, and the preprocessing procedure at least includes background noise removal, gain processing, filtering processing and data zero point correction; The data zero point correction is performed by searching for a specific signal feature of each radar data channel within a preset starting time window, determining an actual signal zero point of each radar data channel, cutting or zeroing the data of a sampling point before the actual signal zero point, and performing time offset and resampling on the remaining data, so that the signal zero points of all radar data channels are aligned in the time dimension, and a radar data body with consistent depth direction is generated. 3.The multi-dimensional presentation method based on intelligent analysis of three-dimensional ground penetrating radar data according to claim 1, characterized in that, The geological radar data sequence carrying spatial position attributes includes: The positioning data is analyzed to obtain the time stamp and latitude and longitude coordinates of each positioning point; The acquisition time stamp of each radar data channel in the original three-dimensional geological radar data is extracted; Each radar data channel is matched with multiple adjacent positioning points based on the time stamp, and the spatial coordinates of each radar data channel are calculated through an interpolation algorithm; The calculated spatial coordinates are associated with the corresponding radar data channel as attribute information to form a geological radar data sequence carrying spatial position attributes. 4.The multi-dimensional presentation method based on intelligent analysis of three-dimensional ground penetrating radar data according to claim 1, characterized in that, The multi-channel vertical profile data set construction process includes: Taking the radar data of each detection channel as input, the image width is determined according to the number of channels, and the image height is determined according to the number of sampling points, and the pixel grid of the two-dimensional image is constructed; Single-column pixel mapping is performed on each radar data channel, and the amplitude value sequence of each radar data channel is matched to the vertical axis coordinate position of the corresponding image column pixel one by one based on the amplitude value acquisition sequence of the data channel; Each sampling point amplitude value is independently converted into a corresponding RGB color value through a predetermined color mapping mechanism, and filled into the mapped pixel position, thereby generating a vertical profile image of each channel, and the color mapping mechanism includes normalizing the amplitude value and applying a pseudo-color or grayscale mapping table.

5. The method for multi-dimensional presentation based on intelligent analysis of three-dimensional GPR data according to claim 4, characterized in that, The multi-depth horizontal slice image data set construction process includes: The actual depth corresponding to each sampling point is determined according to the radar wave speed and sampling frequency parameters; For each target depth, the pixel row corresponding to the depth is extracted from the vertical profile image of all channels; The extracted pixel rows are spliced according to the physical space position of the corresponding channel to generate a horizontal slice image representing the structure distribution of the depth level. 6.The multi-dimensional presentation method based on intelligent analysis of three-dimensional ground penetrating radar data according to claim 5, characterized in that, The mapping relationship between the image pixels and the actual physical size includes: The physical length value corresponding to the pixels of the image in the detection direction is determined based on the channel spacing; The physical depth value corresponding to the pixels of the image in the depth direction is determined based on the sampling frequency, radar wave speed and number of sampling points; The physical width value corresponding to the pixels of the image in the width direction is determined based on the number of channels and the channel spacing; According to the determined physical size mapping relationship, corresponding physical scales are generated and added to the vertical profile image and the horizontal slice image. 7.The multi-dimensional presentation method based on intelligent analysis of three-dimensional GPR data according to claim 1, wherein, The multi-dimensional collaborative visualization interface is constructed, including: A vertical profile display area, a horizontal slice display area, and a video playback area are set up to display the selected vertical profile image, the horizontal slice image at a specific depth level, and the synchronous video frame, respectively; A time-space linkage mechanism among the three areas is established to synchronously update the display content of the remaining areas in response to user interaction in any area; The linkage mechanism specifically implements the following synchronization: i. According to the vertical position selected on the vertical profile image, the horizontal slice display area is synchronously updated to the slice at the corresponding depth, and the video playback area is controlled to jump to the video frame at the corresponding time; ii. According to the horizontal position selected on the horizontal slice image, the vertical profile display area is synchronously switched to the vertical profile passing through the position, and the video playback area is controlled to jump to the video frame at the corresponding time; iii. According to the adjustment of the video playback progress, the vertical profile and the horizontal slice display area are synchronously updated to the radar image corresponding to the time and position of the current video frame. The physical size calculation process includes: Based on the abnormal body boundary information output by the target recognition model, the recognized abnormal body is outlined; 8.The multi-dimensional presentation method based on intelligent analysis of three-dimensional GPR data according to claim 1, wherein, The number of pixels occupied by the outlined area in the vertical profile image and the horizontal slice image along the length, width, and depth directions is respectively counted; Combined with the pre-established mapping relationship between image pixels and actual physical size, the pixel number is multiplied by the corresponding unit pixel physical value to calculate the actual length, actual width, and actual depth of the abnormal body. It includes: A data acquisition module acquires original three-dimensional geological radar data, synchronous video data, and positioning data; 9. A multi-dimensional display system based on intelligent analysis of three-dimensional ground penetrating radar data, characterized in that, A radar data processing module pre-processes the original three-dimensional geological radar data and assigns spatial coordinate information in combination with the positioning data to generate a geological radar data sequence carrying spatial position attributes; A video data alignment module performs timestamp alignment processing on the synchronous video data to generate a video frame sequence that is time-synchronized with the geological radar data sequence; An image conversion module constructs a multi-channel vertical profile image dataset and a multi-depth horizontal slice image dataset based on the geological radar data sequence, and establishes a mapping relationship between image pixels and actual physical size; A multi-dimensional display module constructs a multi-dimensional collaborative visualization interface that synchronously displays the multi-channel vertical profile image, the multi-depth horizontal slice image, and the synchronous video frame in the same view, and supports linkage response based on timestamps and spatial coordinates; An intelligent recognition module performs intelligent fusion analysis on the multi-channel vertical profile image, the multi-depth horizontal slice image, and the synchronous video frame, identifies geological abnormal bodies, and calculates their physical sizes; A detection output module outputs a detection report containing abnormal body physical size and spatial position information, and labels the three-dimensional position of the abnormal body in the geographic space platform. ​ ​

Images