Training method of three-dimensional ground penetrating radar underground target recognition network

Abstract

The invention provides a training method of a three-dimensional ground penetrating radar underground target recognition network, and relates to the technical field of underground survey and nondestructive testing, and the training method of the underground target recognition network comprises the steps: obtaining three-dimensional ground penetrating radar original data containing a target object; analyzing data characteristics of the three-dimensional ground penetrating radar original data, and obtaining two-dimensional ground penetrating radar slice data corresponding to a horizontal plane, a vertical section and a cross section of the target object; expanding the two-dimensional ground penetrating radar slice data based on a cyclic generative adversarial network, and merging the expanded two-dimensional ground penetrating radar slice data to generate target-containing three-dimensional ground penetrating radar data; the three-dimensional ground penetrating radar original data and the target-containing three-dimensional ground penetrating radar data are converted into the point cloud data, and the point cloud data are utilized to train the preset recognition network to obtain the target recognition network, so that the underground target can be quickly, efficiently and accurately recognized, the target recognition efficiency is improved, and the false detection condition is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of underground survey and nondestructive testing, and in particular to a training method for an underground target recognition network of a three-dimensional ground penetrating radar. Background Art

[0002] With the frequent occurrence of road disasters across the country, in order to protect the lives and property of the people, it is urgent to conduct regular and thorough safety hazard inspections on highways. However, my country's roads are widely covered, with a huge total mileage and diverse road conditions. Traditional road inspection methods are time-consuming and labor-intensive, and it is difficult to conduct all-round and full-coverage inspections. As a non-destructive inspection method with high efficiency, large coverage area, rich data information, high accuracy and strong reliability, 3D ground penetrating radar is considered to be the best inspection method for detecting underground hazards on the road surface and inside the road.

[0003] However, it is difficult to obtain field measured data containing road diseases, which cannot meet the data scale required by deep learning, which limits the possibility of many methods in practical engineering applications. Therefore, how to make full use of the existing small amount of 3D ground penetrating radar data to effectively improve the accuracy of target recognition and reduce false detection has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to solve the key problems encountered in the existing three-dimensional ground penetrating radar target recognition technology, especially when using deep learning, the amount of three-dimensional ground penetrating radar data is insufficient and the spatial correlation of three-dimensional ground penetrating radar data is not fully considered, which in turn affects the accuracy of target recognition. The present invention provides a training method for a three-dimensional ground penetrating radar underground target recognition network.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] A first aspect of an embodiment of the present application provides a training method for a three-dimensional ground penetrating radar underground target recognition network, comprising:

[0007] Acquire three-dimensional ground penetrating radar raw data containing the target object;

[0008] Analyze the data characteristics of the three-dimensional ground penetrating radar raw data to obtain the two-dimensional ground penetrating radar slice data corresponding to the horizontal plane, longitudinal section and cross section of the target object respectively;

[0009] Expanding the two-dimensional ground penetrating radar slice data based on a recurrent generative adversarial network, and merging the expanded two-dimensional ground penetrating radar slice data to generate target three-dimensional ground penetrating radar data;

[0010] The three-dimensional ground penetrating radar raw data and the target three-dimensional ground penetrating radar data are converted into point cloud data, and the point cloud data are used to train a preset recognition network to obtain a target recognition network.

[0011] Optionally, after acquiring the three-dimensional ground penetrating radar raw data containing the target object, the method further includes:

[0012] The initial three-dimensional ground penetrating radar data is preprocessed and calibrated; the preprocessing includes removing noise, removing DC drift, removing background and / or adjusting gain; the data calibration includes time zero point calibration and / or amplitude calibration.

[0013] Optionally, the data characteristics include amplitude, phase and / or frequency; analyzing the data characteristics of the three-dimensional ground penetrating radar raw data to obtain two-dimensional ground penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section respectively, includes:

[0014] Analyzing the amplitude, phase and / or frequency of the three-dimensional ground penetrating radar raw data to obtain the reflection coefficient between the target object and the background medium;

[0015] Based on the reflection coefficient, two-dimensional ground penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section are obtained.

[0016] Optionally, merging the expanded two-dimensional ground penetrating radar slice data to generate three-dimensional ground penetrating radar data containing the target includes:

[0017] Acquire data information of each channel of the three-dimensional ground penetrating radar; the data information includes the antenna channel number;

[0018] The expanded two-dimensional ground penetrating radar slice data are merged according to the antenna channel number to generate three-dimensional ground penetrating radar data containing the target.

[0019] Optionally, the preset recognition network is a PointNet++ network.

[0020] A second aspect of the embodiment of the present application provides a training device for a three-dimensional ground penetrating radar underground target recognition network, comprising: an acquisition module, an analysis module, an expansion module and a training module, wherein:

[0021] The acquisition module is configured to acquire the three-dimensional ground penetrating radar raw data containing the target object;

[0022] The analysis module is configured to analyze the data characteristics of the three-dimensional ground penetrating radar raw data to obtain the two-dimensional ground penetrating radar slice data corresponding to the horizontal plane, longitudinal section and cross section of the target object respectively;

[0023] The expansion module is configured to expand the two-dimensional ground penetrating radar slice data based on a cyclic generative adversarial network, and merge the expanded two-dimensional ground penetrating radar slice data to generate target three-dimensional ground penetrating radar data;

[0024] The training module is configured to convert the three-dimensional ground penetrating radar raw data and the three-dimensional ground penetrating radar data containing the target into point cloud data, and use the point cloud data to train a preset recognition network to obtain a target recognition network.

[0025] Optionally, the device further includes a data processing module; the data processing module is configured to:

[0026] The three-dimensional ground penetrating radar raw data is preprocessed and calibrated; the preprocessing includes removing noise, removing DC drift, removing background and / or adjusting gain; the data calibration includes time zero point calibration and / or amplitude calibration.

[0027] Optionally, the data characteristics include amplitude, phase and / or frequency; and the analysis module is specifically configured as follows:

[0028] Analyzing the amplitude, phase and / or frequency of the three-dimensional ground penetrating radar raw data to obtain the reflection coefficient between the target object and the background medium;

[0029] Based on the reflection coefficient, two-dimensional ground penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section are obtained.

[0030] A third aspect of an embodiment of the present application provides an electronic device, including a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the training method for the three-dimensional ground penetrating radar underground target recognition network described in the first aspect.

[0031] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0032] Compared with the prior art, the technical solution provided by this application has the following beneficial effects:

[0033] The present invention provides a training method for a three-dimensional ground-penetrating radar underground target recognition network, by acquiring three-dimensional ground-penetrating radar raw data containing a target object, analyzing the data characteristics of the three-dimensional ground-penetrating radar raw data, acquiring two-dimensional ground-penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section respectively, and expanding the two-dimensional ground-penetrating radar slice data based on a cyclic generative adversarial network, merging the expanded two-dimensional ground-penetrating radar slice data, generating three-dimensional ground-penetrating radar data containing a target, thereby converting the three-dimensional ground-penetrating radar raw data and the target three-dimensional ground-penetrating radar data into point cloud data, and using the point cloud data to train a preset recognition network to obtain a target recognition network. In the case of scarcity of three-dimensional ground-penetrating radar data containing a target, the present invention combines the data characteristics of the three-dimensional ground-penetrating radar data to perform data conversion, and uses a cyclic generative adversarial network to expand data samples, making maximum use of the rich information provided by the three-dimensional data, and using a three-dimensional point cloud target detection algorithm for feature extraction, and finally realizing fast, efficient and accurate underground target recognition, improving the accuracy of target recognition, and reducing the situation of false detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flowchart of a method for training a three-dimensional ground penetrating radar underground target recognition network provided in an embodiment of the present application;

[0035] Figure 2 A schematic diagram of the structure of a training device for a three-dimensional ground penetrating radar underground target recognition network provided in an embodiment of the present application;

[0036] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] Below, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application.

[0038] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "include", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0039] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0040] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flow charts may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions, when executed by the processor, may create a device for implementing the functions / operations described in these block diagrams and / or flow charts.

[0041] In some embodiments, see Figure 1 , Figure 1 A flow chart of a training method for a three-dimensional ground penetrating radar underground target recognition network provided in an embodiment of the present application; a training method for a three-dimensional ground penetrating radar underground target recognition network provided in an embodiment of the present application includes:

[0042] S110, obtaining three-dimensional ground penetrating radar raw data containing the target object.

[0043] Here, the target object can be an underground cavity, a manhole cover, or a pipeline. It should be noted that after collecting the 3D GPR data, the abnormal location can be determined based on the recognition result, and a hole can be opened for verification. The actual hole opening verification information can be compared and matched with the 3D GPR raw data to verify and calibrate the 3D GPR raw data, ensuring that the selected 3D GPR raw data contains the target object.

[0044] In some embodiments, S110, after acquiring the three-dimensional ground penetrating radar raw data containing the target object, the training method of the underground target recognition network further includes:

[0045] The three-dimensional ground penetrating radar raw data is preprocessed and calibrated; the preprocessing includes removing noise, removing DC drift, removing background and / or adjusting gain; the data calibration includes time zero point calibration and / or amplitude calibration.

[0046] Noise removal refers to removing high-frequency noise and random noise from the data through filtering techniques (such as bandpass filtering, Wiener filtering, etc.). Removing DC drift refers to eliminating DC offset caused by antenna coupling or environmental factors. Removing background includes subtracting the average value of all data in each GPR data segment. Adjusting gain refers to adjusting the gain according to the dynamic range of the data to better display the underground structure. Time zero calibration is used to ensure that the time zero point of the GPR echo is correctly aligned. Amplitude calibration is to calibrate the amplitude according to the characteristics of the GPR system to accurately reflect the electrical properties of the underground medium. It should be noted that the preprocessing and data calibration operations can also include other methods, such as removing direct waves, that is, setting a threshold according to the specific parameters of the GPR, and then directly truncating the data within the threshold range to remove the direct wave signal. By preprocessing and calibrating the raw data of the 3D GPR, data errors can be reduced and the accuracy and reliability of the final model training can be ensured.

[0047] In one example, the gain is adjusted using the following gain processing mathematical model:

[0048] y'(n)=y(n)·h(n)

[0049] Among them, y(n) is the data before gain processing, h(n) is the gain function, and y'(n) is the data after gain processing.

[0050] h(n)=[f(1),f(2),…,f(M)]

[0051] Among them, f(i)=i a (i=1,2,…,M), a is an exponential factor used to adjust the amplitude of each element in the gain vector, and the value of a can be 0.7.

[0052] S120, analyzing the data characteristics of the 3D ground penetrating radar raw data, and obtaining the 2D ground penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section respectively.

[0053] In this embodiment, it is first necessary to clarify the specific type of the target object, such as underground cavities, manhole covers, pipelines, etc. According to the characteristics of the target object, the data area that may contain the target can be preliminarily located in the three-dimensional ground penetrating radar data. By using image processing software (such as 3dr-Examiner, etc.) to perform data cutting, the three-dimensional data body containing the target can be extracted. The two-dimensional ground penetrating radar slice of the target object can be obtained from the three-dimensional data body. Among them, the horizontal plane shows the target distribution on the ground plane. On the horizontal plane, the plane distribution and relative position relationship of the target can be observed. The cross section shows the profile information perpendicular to the ground direction. The longitudinal section shows the profile information along the ground direction. On the longitudinal section, the vertical distribution and burial depth of the target can be observed. On the cross section, the horizontal distribution and morphological characteristics of the target object, such as the length, width, shape, etc. of the target object, can be observed.

[0054] In some embodiments, the data characteristics include amplitude, phase and / or frequency; S120, analyzing the data characteristics of the 3D ground penetrating radar raw data to obtain 2D ground penetrating radar slice data corresponding to the horizontal plane, longitudinal section and cross section of the target object, including:

[0055] Analyze the amplitude, phase and / or frequency of the 3D ground penetrating radar raw data to obtain the reflection coefficient between the target object and the background medium;

[0056] Based on the reflection coefficient, the two-dimensional ground penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section are obtained.

[0057] By analyzing the characteristics of the reflected waves on the horizontal plane, the type and quantity of the target as well as possible interferences can be identified; by analyzing the characteristics of the reflected waves on the longitudinal section, the continuity and integrity of the target as well as possible faults or cracks can be determined; by analyzing the characteristics of the reflected waves on the cross section, such as amplitude and phase, the physical properties of the target, such as density and water content, can be inferred.

[0058] In this embodiment, a fixed depth is selected in the three-dimensional data volume, and the reflection information is extracted in the horizontal direction to generate a horizontal slice. The horizontal slice shows the distribution of the target object in the horizontal plane at the depth. A fixed horizontal position is selected, and the reflection information is extracted in the vertical direction to generate a longitudinal slice. The longitudinal slice shows the distribution of the target object in the vertical direction at the position. A fixed depth range is selected, and the reflection information is extracted in the horizontal direction to generate a cross-sectional slice. The cross-sectional slice shows the distribution of the target object in the lateral direction within the depth range.

[0059] S130, expanding the two-dimensional ground penetrating radar slice data based on a cyclic generative adversarial network, and merging the expanded two-dimensional ground penetrating radar slice data to generate three-dimensional ground penetrating radar data containing the target.

[0060] In this embodiment, based on the two-dimensional ground penetrating radar slices, data sets containing targets can be constructed respectively, and each data set includes two sub-data sets, namely, data containing targets and soil background data without targets. By inputting the data set into the pre-created recurrent generative adversarial network, the two-dimensional radar slice data is expanded. On this basis, the analysis results of the cross section, longitudinal section and horizontal plane are integrated to form more comprehensive target three-dimensional ground penetrating radar data.

[0061] In some embodiments, S130, merging the expanded two-dimensional ground penetrating radar slice data to generate target three-dimensional ground penetrating radar data includes:

[0062] Obtain data information of each channel of the three-dimensional ground penetrating radar; the data information includes the antenna channel number;

[0063] The expanded two-dimensional ground penetrating radar slice data are merged according to the antenna channel number to generate the target three-dimensional ground penetrating radar data.

[0064] In this embodiment, according to the working principle and system design of the ground penetrating radar, the order and correspondence between different antenna channels are determined to ensure that each two-dimensional slice data corresponds to the correct antenna channel number, so as to realize the alignment processing of the two-dimensional slice data, and then place each two-dimensional slice into the three-dimensional space according to its corresponding depth or position, complete the merging of the two-dimensional radar slice data, and generate the target three-dimensional ground penetrating radar data.

[0065] S140, converting the 3D ground penetrating radar raw data and the 3D ground penetrating radar target data into point cloud data, and using the point cloud data to train a preset recognition network to obtain a target recognition network.

[0066] In this embodiment, the three-dimensional ground penetrating radar raw data and the three-dimensional ground penetrating radar target data are converted into point cloud data containing position information and reflection intensity information through coordinate transformation. The point cloud data will contain a large number of three-dimensional discrete points, each of which represents the information of a certain position detected by the ground penetrating radar. The point cloud data and the extracted features are used as input to train the preset recognition network. During the training process, the parameters of the network (such as learning rate, number of iterations, etc.) are continuously adjusted, and the validation set can be used to evaluate and optimize the preset recognition network to improve the accuracy and robustness of target recognition until the requirements are met and a stable target recognition network is obtained.

[0067] In some embodiments, the preset recognition network is a PointNet++ network.

[0068] PointNet++ is an extension based on PointNet, which aims to process point cloud data hierarchically, from local to global, and gradually abstract the spatial structure and semantic information in the point cloud. It realizes tasks such as classification, segmentation and recognition of 3D point clouds by extracting features in a hierarchical manner. Before network training, point cloud data can also be enhanced, such as scaling, translation, rotation and other operations. After introducing reflection intensity as an additional attribute of point cloud data based on position information, its reflection intensity attribute will not be changed during data enhancement. Therefore, by integrating intensity information at each point, the point cloud can be more comprehensively represented and feature learning can be better achieved.

[0069] In one example, the training parameters of the PointNet++ network training phase are set as follows: the batch size is set to 8, the learning rate is 1×10 -4 , the total number of training epochs is set to 500, and the weight decay is 1×10 -4 , the back-propagation optimizer Adam is used to optimize the training speed and convergence effect. In addition, the number of points in each target is inconsistent. In order to ensure that the input format of the network is consistent when batch training different point cloud data, all point cloud data needs to be sampled before training. The number of points of each target after sampling is 46080.

[0070] In one example, the ground penetrating radar device is a fixed frequency ground penetrating radar device, and the fixed frequency is 200 MHz and 450 MHz.

[0071] In another example, the size of the three-dimensional data volume is 256 pixels × 256 pixels × the number of antenna array channels. In the process of denoising, the bandpass filtering process specifically includes designing an FIR bandpass filter, the filter type is equal ripple, the stopband attenuation is 40 dB, the passband gain is 0 dB, and the upper limit of the passband ripple oscillation is 1 dB.

[0072] In the case of scarcity of three-dimensional ground penetrating radar data containing targets, this embodiment combines the data characteristics of the three-dimensional ground penetrating radar data to perform data conversion, and uses a cyclic generative adversarial network to expand data samples to maximize the use of the rich information provided by the three-dimensional data; at the same time, a three-dimensional point cloud target detection algorithm is used for feature extraction, and ultimately fast, efficient, and accurate underground target recognition is achieved, which greatly improves the efficiency and accuracy of target recognition and reduces the false detection rate.

[0073] In some embodiments, see Figure 2 , Figure 2A schematic diagram of the structure of a training device for a three-dimensional ground penetrating radar underground target recognition network provided in an embodiment of the present application; a training device 200 for a three-dimensional ground penetrating radar underground target recognition network provided in an embodiment of the present application comprises: an acquisition module 210, an analysis module 220, an expansion module 230 and a training module 240, wherein the acquisition module 210 is configured to acquire initial three-dimensional ground penetrating radar data containing a target object;

[0074] The analysis module 220 is configured to analyze the data characteristics of the 3D GPR raw data and obtain the 2D GPR slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section respectively;

[0075] An expansion module 230 is configured to expand the two-dimensional ground penetrating radar slice data based on a cyclic generative adversarial network, and merge the expanded two-dimensional radar slice data to generate target three-dimensional ground penetrating radar data;

[0076] The training module 240 is configured to convert the 3D GPR raw data and the 3D GPR target data into point cloud data, and use the point cloud data to train a preset recognition network to obtain a target recognition network.

[0077] In some embodiments, the underground target recognition network training device 200 further includes a data processing module; the data processing module is configured to:

[0078] The three-dimensional ground penetrating radar raw data is preprocessed and calibrated; the preprocessing includes removing noise, removing DC drift, removing background and / or adjusting gain; the data calibration includes time zero point calibration and / or amplitude calibration.

[0079] In some embodiments, the data characteristics include amplitude, phase and / or frequency; the analysis module 220 is specifically configured to:

[0080] Analyze the amplitude, phase and / or frequency of the 3D ground penetrating radar raw data to obtain the reflection coefficient between the target object and the background medium;

[0081] Based on the reflection coefficient, two-dimensional radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section are obtained.

[0082] In some embodiments, the expansion module 230 is specifically configured as follows:

[0083] Obtain data information of each channel of the three-dimensional ground penetrating radar; the data information includes the antenna channel number;

[0084] The expanded two-dimensional ground-penetrating radar slice data are merged according to the antenna channel number to generate three-dimensional ground-penetrating radar data containing the target.

[0085] In some embodiments, the preset recognition network is a PointNet++ network.

[0086] The training device for the three-dimensional ground penetrating radar underground target recognition network provided in the embodiment of the present application is based on the above-mentioned training method for the three-dimensional ground penetrating radar underground target recognition network. To avoid repetition, it will not be repeated here.

[0087] It should be noted that the training device for the three-dimensional ground penetrating radar underground target recognition network provided in the embodiment of the present application and the training method for the three-dimensional ground penetrating radar underground target recognition network provided in the embodiment of the present application are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned underground target recognition network training method, and the repeated parts will not be repeated.

[0088] In some embodiments, see Figure 3 , Figure 3 The electronic device 300 provided in the embodiment of the present application includes a processor 310 and a memory 320; the memory 320 stores a computer program, wherein the computer program implements the above-mentioned training method of the underground target recognition network of the three-dimensional ground penetrating radar when executed by the processor.

[0089] Specifically, the processor 310 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 310 may also include an onboard memory for cache purposes. The processor 310 may be a single processing unit or multiple processing units for executing different actions of the method flow according to the embodiment of the present application.

[0090] The memory 320 may be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, the memory 320 may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device, component or propagation medium. Specific examples of the memory 320 include: a magnetic storage device, such as a magnetic tape or a hard disk (HDD); an optical storage device, such as a compact disk (CD-ROM); a random access memory (RAM) or flash memory; and / or a wired / wireless communication link.

[0091] The present application also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned training method for an underground target recognition network. The computer-readable medium may be included in the device / apparatus / system described in the above-mentioned embodiments; or it may exist independently without being assembled into the device / apparatus / system. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed, the method according to the embodiment of the present application is implemented.

[0092] According to an embodiment of the present application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, optical cable, radio frequency signal, etc., or any suitable combination of the above.

[0093] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present application. Therefore, the scope of the present application should not be limited to the above-described embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

Claims

1. A training method for a three-dimensional ground penetrating radar underground target recognition network, characterized in that: include: Acquire three-dimensional ground penetrating radar raw data containing the target object; Analyze the data characteristics of the three-dimensional ground penetrating radar raw data to obtain the two-dimensional ground penetrating radar slice data corresponding to the horizontal plane, longitudinal section and cross section of the target object respectively; Expanding the two-dimensional ground penetrating radar slice data based on a recurrent generative adversarial network, and merging the expanded two-dimensional ground penetrating radar slice data to generate three-dimensional ground penetrating radar data containing the target; The three-dimensional ground penetrating radar raw data and the three-dimensional ground penetrating radar data containing the target are converted into point cloud data, and the point cloud data are used to train a preset recognition network to obtain a target recognition network.

2. The training method of underground target recognition network according to claim 1, characterized in that: After acquiring the three-dimensional ground penetrating radar raw data containing the target object, the method further includes: The three-dimensional ground penetrating radar raw data is preprocessed and calibrated; the preprocessing includes removing noise, removing DC drift, removing background and / or adjusting gain; the data calibration includes time zero point calibration and / or amplitude calibration.

3. The training method of underground target recognition network according to claim 1, characterized in that: The data characteristics include amplitude, phase and / or frequency; analyzing the data characteristics of the three-dimensional ground penetrating radar raw data to obtain two-dimensional ground penetrating radar slice data corresponding to the horizontal plane, longitudinal section and cross section of the target object, including: Analyzing the amplitude, phase and / or frequency of the three-dimensional ground penetrating radar raw data to obtain the reflection coefficient between the target object and the background medium; Based on the reflection coefficient, two-dimensional ground penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section are obtained.

4. The training method for underground target recognition network according to claim 1, characterized in that: The method of merging the expanded two-dimensional ground penetrating radar slice data to generate three-dimensional ground penetrating radar data containing the target includes: Acquire data information of each channel of the three-dimensional ground penetrating radar; the data information includes the antenna channel number; The expanded two-dimensional ground penetrating radar slice data are merged according to the antenna channel number to generate three-dimensional ground penetrating radar data containing the target.

5. The training method of the three-dimensional ground penetrating radar underground target recognition network according to claim 1, characterized in that: The preset recognition network is a PointNet++ network.

6. A training device for a three-dimensional ground penetrating radar underground target recognition network, characterized in that: include: Acquisition module, analysis module, expansion module and training module, where: The acquisition module is configured to acquire the three-dimensional ground penetrating radar raw data containing the target object; The analysis module is configured to analyze the data characteristics of the three-dimensional ground penetrating radar raw data to obtain the two-dimensional ground penetrating radar slice data corresponding to the horizontal plane, longitudinal section and cross section of the target object respectively; The expansion module is configured to expand the two-dimensional ground penetrating radar slice data based on a cyclic generative adversarial network, and merge the expanded two-dimensional ground penetrating radar slice data to generate three-dimensional ground penetrating radar data containing the target; The training module is configured to convert the three-dimensional ground penetrating radar raw data and the three-dimensional ground penetrating radar data containing the target into point cloud data, and use the point cloud data to train a preset recognition network to obtain a target recognition network.

7. The training device for the three-dimensional ground penetrating radar underground target recognition network according to claim 6, characterized in that: The device further includes a data processing module; the data processing module is configured to: The three-dimensional ground penetrating radar raw data is preprocessed and calibrated; the preprocessing includes removing noise, removing DC drift, removing background and / or adjusting gain; the data calibration includes time zero point calibration and / or amplitude calibration.

8. The training device for the three-dimensional ground penetrating radar underground target recognition network according to claim 6, characterized in that: The data characteristics include amplitude, phase and / or frequency; the analysis module is specifically configured as follows: Analyzing the amplitude, phase and / or frequency of the three-dimensional ground penetrating radar raw data to obtain the reflection coefficient between the target object and the background medium; Based on the reflection coefficient, two-dimensional ground penetrating radar slice data corresponding to the target object in the horizontal plane, longitudinal section and cross section are obtained.

9. An electronic device comprising a processor and a memory; the memory stores a computer program, wherein: When the computer program is executed by the processor, the computer program implements the training method for an underground target recognition network according to any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images