Underground cavity automatic identification method based on three-dimensional ground penetrating radar data

Abstract

The invention discloses an underground cavity automatic identification method based on three-dimensional ground penetrating radar data, and the method comprises the following steps: S1, carrying out the signal preprocessing of the collected three-dimensional ground penetrating radar data, segmenting the data, and obtaining a processing unit; s2, performing forward reasoning on data of each processing unit by using a deep learning algorithm; and S3, carrying out position interpretation on the reasoning result to obtain a hole identification result. The method comprises the steps of performing signal preprocessing on acquired three-dimensional ground penetrating radar data, segmenting the data to obtain processing units, performing forward reasoning on the data of each processing unit by using a deep learning algorithm, extracting features by using a 3D feature network, performing detection by using spatial dimensionality reduction to obtain 2D features, and finally performing position interpretation on a reasoning result. According to the method and the device, the void recognition result is obtained, so that the information between the channels of the three-dimensional ground penetrating radar is fused, supervision training can be carried out by using 2D target frame labeling, and a better recognition effect is obtained under the condition of relatively low cost.

Description

Technical Field

[0001] This invention relates to the field of underground cavity identification technology, and in particular to an automatic identification method for underground cavities based on three-dimensional ground-penetrating radar data. Background Technology

[0002] With the continuous development of urban infrastructure and the increasing complexity of underground pipelines, 3D ground-penetrating radar (GPR) is widely used for identifying underground cavities in municipal roads due to its advantages such as non-destructive operation, strong imaging capabilities, and wide adaptability. Traditional GPR systems mainly focus on the acquisition and analysis of two-dimensional data, which is insufficient to meet the needs for three-dimensional reconstruction of spatial structures in complex scenarios. Therefore, 3D GPR systems have gradually become a hot topic in research and engineering applications.

[0003] Existing methods for automatic identification of underground cavities using deep learning algorithms based on 3D ground-penetrating radar data generally fall into two categories: one is based on B-Scan data and uses 2D target detection algorithms to obtain underground cavity targets. The problem with this method is that it cannot fuse features between 3D radar channels based on single-channel data, leading to a high number of false detections. The other is based on C-Scan data and uses 3D target detection algorithms to obtain underground cavity targets. This method has a high cost in obtaining 3D annotations during the model training phase, and the 3D target bounding boxes cannot accurately represent the shape and location of underground cavity targets, resulting in poor algorithm implementation performance. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data.

[0005] The objective of this invention is achieved through the following technical solution: a method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data, comprising the following steps:

[0006] S1: Perform signal preprocessing on the acquired 3D ground-penetrating radar data and segment the data to obtain processing units;

[0007] S2: Perform forward inference on the data of each processing unit using a deep learning algorithm;

[0008] S3: Perform positional interpretation on the inference results to obtain the hole identification results.

[0009] Preferably, step S1 further includes the following step:

[0010] S11: Perform signal preprocessing on 3D ground-penetrating radar data.

[0011] ;

[0012] in, This is the raw signal data of a three-dimensional ground-penetrating radar. For radar signal preprocessing, To complete the data after signal preprocessing, This refers to the number of channels of the 3D ground-penetrating radar. This represents the number of samples taken at the depth of the three-dimensional ground-penetrating radar. This represents the number of measurement points in each channel along the measurement direction.

[0013] S12: Based on the measurement point location information, the measurement points are intercepted at equal intervals, and the 3D ground-penetrating radar data is divided into C-Scan processing units.

[0014] ;

[0015] in, For the first One C-Scan data point.

[0016] Preferably, in step S11, signal preprocessing includes zero bias correction, zero-point adjustment, background field removal, filtering and noise reduction, gain adjustment, and contrast adjustment.

[0017] Preferably, step S2 further includes the following step:

[0018] S21: Use a 3D convolutional neural network to extract features from C-Scan data.

[0019] ;

[0020] in, This is a multi-scale feature extraction algorithm. This represents a feature at one of the scales;

[0021] S22: Multi-scale feature fusion using a cross-scale feature fusion network.

[0022] ;

[0023] in, For multi-layer feature fusion transformation, Enhanced 3D features after fusion;

[0024] S23: Use spatial pooling networks to reduce the dimensionality of feature tensors in 3D space to 2D space.

[0025] ;

[0026] in, For spatial pooling dimensionality reduction transformation, These are the 2D features after dimensionality reduction.

[0027] S24: Use a 2D-anchor-free detection network to perform position parameter regression on 2D features.

[0028] ;

[0029] in, This represents a 2D detection network.

[0030] Preferably, in step S24, It is in the size of A 6-channel tensor on the grid, where the first channel represents the confidence level of the target corresponding to that grid, and the second to fifth channels are descriptions of the 2D profile target bounding boxes. The last channel data represents the channel position corresponding to the grid target.

[0031] .

[0032] Preferably, step S3 further includes the following step:

[0033] S31: Perform profile position interpretation on the output of step S24 to obtain the target box coordinates of the hole in the profile.

[0034] ;

[0035] in, To interpret changes in position, This indicates the location of the cavity in the cross-section.

[0036] S32: Perform channel position interpretation on the output of step S24 to obtain the channel position where the profile target box is located.

[0037] ;

[0038] in For channel interpretation changes, for The location of the passage corresponding to the cavity.

[0039] The present invention has the following advantages: By preprocessing the acquired 3D ground-penetrating radar data and segmenting the data to obtain processing units, the present invention performs forward inference on the data of each processing unit using a deep learning algorithm, thereby extracting features using a 3D feature network, and then using spatial dimensionality reduction to 2D features for detection. Finally, the inference results are interpreted to obtain the hole recognition result. Thus, it not only integrates the information between 3D ground-penetrating radar channels, but also uses 2D target box annotation for supervised training, achieving better recognition results at a lower cost. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the process flow for an automatic identification method for underground cavities. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0042] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0043] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0044] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0045] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0046] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0047] In this embodiment, as Figure 1 As shown, a method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data includes the following steps:

[0048] S1: Perform signal preprocessing on the acquired 3D ground-penetrating radar data and segment the data to obtain processing units;

[0049] S2: Perform forward inference on the data of each processing unit using a deep learning algorithm;

[0050] S3: The inference results are interpreted to obtain the cavity identification results. By preprocessing the acquired 3D ground-penetrating radar data and segmenting the data into processing units, a deep learning algorithm is used for forward inference on each processing unit. This allows for feature extraction using a 3D feature network, followed by spatial dimensionality reduction to 2D features for detection. Finally, the inference results are interpreted to obtain the cavity identification results. This approach not only integrates information from different 3D ground-penetrating radar channels but also utilizes 2D bounding box annotations for supervised training, achieving better recognition results at a lower cost.

[0051] Furthermore, step S1 also includes the following steps:

[0052] S11: Perform signal preprocessing on 3D ground-penetrating radar data.

[0053] ;

[0054] in, This is the raw signal data of a three-dimensional ground-penetrating radar. For radar signal preprocessing, To complete the data after signal preprocessing, This refers to the number of channels of the 3D ground-penetrating radar. This represents the number of samples taken at the depth of the three-dimensional ground-penetrating radar. This represents the number of measurement points in each channel along the measurement direction.

[0055] S12: Based on the measurement point location information, the measurement points are intercepted at equal intervals, and the 3D ground-penetrating radar data is divided into C-Scan processing units.

[0056] ;

[0057] in, For the first The signal preprocessing process involves several C-Scan data points. Further, in step S11, signal preprocessing includes zero-bias correction, zero-point adjustment, background field removal, filtering and noise reduction, gain adjustment, and contrast adjustment. Specifically, each step of the signal preprocessing is implemented using existing methods, and no improvements have been made; therefore, they will not be elaborated upon further.

[0058] In this embodiment, step S2 further includes the following step:

[0059] S21: Use a 3D convolutional neural network to extract features from C-Scan data.

[0060] ;

[0061] in, This is a multi-scale feature extraction algorithm. This represents a feature at one of the scales;

[0062] S22: Multi-scale feature fusion using a cross-scale feature fusion network.

[0063] ;

[0064] in, For multi-layer feature fusion transformation, Enhanced 3D features after fusion;

[0065] S23: Use spatial pooling networks to reduce the dimensionality of feature tensors in 3D space to 2D space.

[0066] ;

[0067] in, For spatial pooling dimensionality reduction transformation, These are the 2D features after dimensionality reduction.

[0068] S24: Use a 2D-anchor-free detection network to perform position parameter regression on 2D features.

[0069] ;

[0070] in, This represents a 2D detection network. Further, in step S24, It is in the size of A 6-channel tensor on the grid, where the first channel represents the confidence level of the target corresponding to that grid, and the second to fifth channels are descriptions of the 2D profile target bounding boxes. The last channel data represents the channel position corresponding to the grid target. It can be seen as the result of splicing the three together in the last dimension.

[0071] .

[0072] In this embodiment, step S3 further includes the following step:

[0073] S31: Perform profile position interpretation on the output of step S24 to obtain the target box coordinates of the hole in the profile.

[0074] ;

[0075] in, To interpret changes in position, This indicates the location of the cavity in the cross-section.

[0076] S32: Perform channel position interpretation on the output of step S24 to obtain the channel position where the profile target box is located.

[0077] ;

[0078] in For channel interpretation changes, for The location of the passage corresponding to the cavity.

[0079] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data, characterized in that: Includes the following steps: S1: Perform signal preprocessing on the acquired 3D ground-penetrating radar data and segment the data to obtain processing units; S2: Perform forward inference on the data of each processing unit using a deep learning algorithm; S3: Perform positional interpretation on the inference results to obtain the hole identification results.

2. The method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data according to claim 1, characterized in that: Step S1 further includes the following steps: S11: Perform signal preprocessing on 3D ground-penetrating radar data. ； in, This is the raw signal data of a three-dimensional ground-penetrating radar. For radar signal preprocessing, To complete the data after signal preprocessing, This refers to the number of channels of the 3D ground-penetrating radar. This represents the number of samples taken at the depth of the three-dimensional ground-penetrating radar. This represents the number of measurement points in each channel along the measurement direction. S12: Based on the measurement point location information, the measurement points are intercepted at equal intervals, and the 3D ground-penetrating radar data is divided into C-Scan processing units. ； in, For the first One C-Scan data point.

3. The method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data according to claim 2, characterized in that: In step S11, signal preprocessing includes zero bias correction, zero-point adjustment, background field removal, filtering and noise reduction, gain adjustment, and contrast adjustment.

4. The method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data according to claim 3, characterized in that: Step S2 further includes the following steps: S21: Use a 3D convolutional neural network to extract features from C-Scan data. ； in, This is a multi-scale feature extraction algorithm. This represents a feature at one of the scales; S22: Multi-scale feature fusion using a cross-scale feature fusion network. ； in, For multi-layer feature fusion transformation, Enhanced 3D features after fusion; S23: Use spatial pooling networks to reduce the dimensionality of feature tensors in 3D space to 2D space. ； in, For spatial pooling dimensionality reduction transformation, These are the 2D features after dimensionality reduction. S24: Use a 2D-anchor-free detection network to perform position parameter regression on 2D features. ； in, This represents a 2D detection network.

5. The method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data according to claim 4, characterized in that: In step S24 It is in the size of A 6-channel tensor on the grid, where the first channel represents the confidence level of the target corresponding to that grid, and the second to fifth channels are descriptions of the 2D profile target bounding boxes. The last channel data represents the channel position corresponding to the grid target. 。 6. The method for automatic identification of underground cavities based on three-dimensional ground-penetrating radar data according to claim 5, characterized in that: Step S3 further includes the following steps: S31: Perform profile position interpretation on the output of step S24 to obtain the target box coordinates of the hole in the profile. ； in, To interpret changes in position, This indicates the location of the cavity in the cross-section. S32: Perform channel position interpretation on the output of step S24 to obtain the channel position where the profile target box is located. ； in For channel interpretation changes, for The location of the passage corresponding to the cavity.

Images