Ground penetrating radar signal denoising method and system based on improved Transform

Abstract

The invention discloses a ground penetrating radar signal denoising method and system based on an improved Transform, relates to the field of signal processing, and solves the problem of low signal-noise separation precision of GPR signals. The method comprises the following steps: acquiring an original GPR signal; performing denoising preprocessing on the original GPR signal to obtain a first GPR signal; the global feature of the first GPR signal is calculated through an improved Transform network; wherein the global feature is obtained through local attention and hyperbola attenuation position coding calculation, and the local attention is obtained through calculation based on the first GPR signal; through an improved Transform network, reconstruction is carried out based on the global features to obtain a second GPR signal; and carrying out smoothing processing and / or island noise removal on the second GPR signal to obtain a target GPR signal. According to the method, the local attention is calculated through the improved Transform architecture network, the global features are calculated in combination with the hyperbolic attenuation position coding, the reflection features of the stratum can be accurately extracted, and differential extraction and reconstruction of the reflection features of the stratums with different depths are realized.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method and system for denoising ground-penetrating radar signals based on an improved Transformer. Background Technology

[0002] In fields such as geological exploration, engineering inspection, and underground pipeline location, ground-penetrating radar (GPR) is a non-destructive detection technology. It transmits high-frequency electromagnetic waves and receives reflected signals from the underground medium to image and analyze underground structures and anomalies. The accuracy of its signal processing directly determines the reliability of the detection results. However, due to the non-homogeneity, multi-interface nature, and complex electromagnetic properties of the underground medium, GPR echo signals are often contaminated by strong noise (including system noise, clutter, and multipath interference), causing target reflection signals (such as pipelines, cavities, and fissures) to be masked. This is especially true for deep, weakly reflected signals, where noise interference can easily cause detection errors or missed detections. Therefore, efficient noise reduction is a core aspect of GPR signal processing. In traditional methods, wavelet transform based on time-frequency analysis and its improved algorithms are widely used. They achieve signal-noise separation through multi-scale decomposition. However, due to the time-frequency resolution of the fixed basis function, it is difficult to match the dynamic multipath effect of electromagnetic wave propagation in underground media. When there are three or more reflection interfaces underground (such as the composite structure of multi-layer soil and pipelines), the composite echo signal formed by multiple reflections of electromagnetic waves will produce severe phase superposition and energy dissipation. Wavelet transform is prone to phase distortion during the decomposition process, resulting in large time positioning errors of the reflected signal. In addition, it has a large amplitude attenuation for deep weak reflection signals, which can easily cause the loss of target features. Therefore, there is a need for a method and system for denoising ground-penetrating radar signals based on an improved Transformer. Summary of the Invention

[0003] To address the issue of low signal-to-noise separation accuracy for GPR signals in existing technologies, this invention provides a ground-penetrating radar (GPR) signal denoising method and system based on an improved Transformer, which can improve the signal-to-noise separation accuracy of GPR signals. The specific technical solution is as follows: In a first aspect, embodiments of this application provide a method for denoising ground-penetrating radar signals based on an improved Transformer, including: The raw Ground Penetrating Radar (GPR) signal is acquired; the raw GPR signal is preprocessed for denoising to obtain a first GPR signal; the global features of the first GPR signal are calculated using an improved Transformer network; wherein the global features are calculated by the improved Transformer network through local attention and hyperbolic attenuation position encoding, and the local attention is calculated by the improved Transformer network based on the first GPR signal; the second GPR signal is reconstructed using the improved Transformer network based on the global features; the second GPR signal is smoothed and / or island noise is removed to obtain the target GPR signal.

[0004] Preferably, the calculation of the global features of the first GPR signal by improving the Transformer network includes: extracting a short-time feature matrix from the first GPR signal based on a preset sliding window; calculating the local attention based on the short-time feature matrix; performing attention enhancement on the short-time feature matrix based on the local attention; and superimposing the attention-enhanced short-time feature matrix and the hyperbolic decay position code to obtain the global features.

[0005] Preferably, before reconstructing the second GPR signal based on the global features, the method further includes: obtaining the relative permittivity, conductivity, and density of the target corresponding to the original GPR signal; constructing a physical constraint code for the target based on the relative permittivity, conductivity, and density; superimposing the physical constraint code and the global features to obtain enhanced global features; the reconstruction of the second GPR signal based on the global features includes: reconstructing the second GPR signal based on the enhanced global features.

[0006] Preferably, the method of reconstructing the second GPR signal based on the global features using the improved Transformer network includes: reconstructing the second GPR signal using the improved Transformer network based on the global features, a reflection guiding mask, and a time-frequency joint constraint loss function; wherein the reflection guiding mask is used to indicate that the attention of the improved Transformer network is focused on the location where geological reflections may occur, and the time-frequency joint constraint loss function is used to optimize the signal reconstruction effect at both the time and frequency domain levels.

[0007] Preferably, the denoising preprocessing of the original GPR signal to obtain the first GPR signal includes: performing a time-varying gain on the original GPR signal based on a logarithmic function; removing the antenna coupling interference component from the time-varying gain original GPR signal based on a preset antenna coupling response template; and normalizing the original GPR signal after removing the antenna coupling interference component to obtain the first GPR signal.

[0008] Preferably, the smoothing and / or islanding noise removal of the second GPR signal includes: smoothing the second GPR signal using a time-varying Gaussian filter.

[0009] Preferably, the smoothing and / or island noise removal of the second GPR signal includes: using dynamic threshold segmentation based on sliding statistics to remove island noise in the second GPR signal.

[0010] Secondly, embodiments of this application provide a ground-penetrating radar signal denoising system based on an improved Transformer, applied to the method described in the first aspect, the system comprising: The acquisition module is used to acquire the raw ground-penetrating radar GPR signal; The first processing module is used to perform noise reduction preprocessing on the original GPR signal to obtain the first GPR signal; The second processing module is used to calculate the global features of the first GPR signal through an improved Transformer network; wherein the global features are calculated by the improved Transformer network through local attention and hyperbolic attenuation position encoding, and the local attention is calculated by the improved Transformer network based on the first GPR signal; The second processing module is also used to reconstruct the second GPR signal based on the global features through the improved Transformer network; The third processing module is used to smooth the second GPR signal and / or remove island noise to obtain the target GPR signal.

[0011] Thirdly, embodiments of this application provide a computing device, including: a memory for storing a program; and a processor for loading the program to execute the method as described in the first aspect.

[0012] Fourthly, embodiments of this application provide a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the method described in the first aspect.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: by improving the local attention calculation of the Transformer architecture network and combining it with the hyperbolic attenuation position encoding to calculate global features, the reflection features of long-range strata can be accurately extracted, and the differentiated extraction and reconstruction of reflection features at different depths can be achieved. Attached Figure Description

[0014] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0015] Figure 1 A flowchart illustrating a ground-penetrating radar signal denoising method based on an improved Transformer, provided for an embodiment of this application; Figure 2 A schematic diagram of a ground-penetrating radar signal denoising system based on an improved Transformer is provided for embodiments of this application; Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0017] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0018] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0019] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0020] To address the issue of low signal-to-noise separation accuracy of GPR signals using traditional techniques, this invention provides a ground-penetrating radar signal denoising method and system based on an improved Transformer, which can improve the signal-to-noise separation accuracy of GPR signals.

[0021] Please see Figure 1 , Figure 1 This application provides a flowchart illustrating a ground-penetrating radar signal denoising method based on an improved Transformer, which is applied to a computing device; as shown below. Figure 1 As shown, the method includes: Step 101: The computing device acquires the raw Ground Penetrating Radar (GPR) signal.

[0022] The computing device can be a personal computer, laptop, smartphone, tablet, or server, or other device with data processing capabilities and the ability to deploy neural network models.

[0023] Among them, GPR signals are records formed by receiving electromagnetic signals reflected or scattered back from the interfaces of underground media after a ground-penetrating radar system transmits high-frequency electromagnetic pulses underground. They are the core information carrier for interpreting underground structures and material distribution. In essence, they are the responses of electromagnetic energy to different media during its underground propagation, containing key information such as the physical properties and structural characteristics of the underground media.

[0024] The computing device can communicate with the GPR signal acquisition device to synchronously acquire the raw GPR signal acquired by the acquisition device; it can also acquire the raw GPR signal that has been acquired from a database or storage device.

[0025] Step 102: The computing device performs noise reduction preprocessing on the original GPR signal to obtain the first GPR signal.

[0026] Preferably, the computing device can perform a time-varying gain on the original GPR signal based on a logarithmic function; then, based on a preset antenna coupling response template, remove the antenna coupling interference component from the original GPR signal after the time-varying gain; and then normalize the original GPR signal after removing the antenna coupling interference component to obtain the first GPR signal.

[0027] The computing device first performs automatic gain control on the input noisy raw GPR signal.

[0028] During propagation, the amplitude of the GPR signal attenuates significantly with depth due to effects such as medium absorption, scattering, and geometric diffusion, severely impacting the detection performance of deep targets. To compensate for this attenuation effect, this method employs time-varying gain control based on a logarithmic function: ; Where t is the acquisition time, representing the signal propagation time in the medium; t0 is the initial time delay compensation parameter, used to avoid numerical instability; and G(t) is the gain amplitude of the GPR signal at time t. This gain function applies a gradual amplification to the early and mid-stage signals, while applying a strong gain to the deep signals, ensuring that the weak deep signals can still maintain sufficient amplitude detectability in subsequent processing stages.

[0029] Then, the computing device can remove the antenna coupling effect component from the GPR signal.

[0030] Near-field coupling exists between the transmitting and receiving antennas of ground-penetrating radar (GPR), causing strong direct-wave interference to the received signal in shallow layers, thus affecting the detection of shallow targets. To eliminate this effect, this method employs deconvolution operations based on measured antenna response templates: ; Where X(t) is the time-domain waveform of the GPR signal after time-varying gain, and h(t) is a pre-determined antenna coupling response template, characterizing the coupling characteristics of the transmit-receive antenna. The deconvolution operation is used to separate the antenna coupling interference component from the received signal; X'(t) is the GPR signal after removing the antenna coupling effect component.

[0031] Specifically, the Antenna Coupling Response Template is a reference model or standard waveform used in fields such as ground-penetrating radar and electromagnetic detection to describe the characteristics of the direct coupling signal between the transmitting and receiving antennas. It mainly reflects the influence of the antenna's own structure, relative position, and surrounding environment (such as air and shallow media) on the original signal, and is a key tool in signal processing for separating "useful target reflected signals" from "antenna direct coupling interference".

[0032] The core logic for distinguishing shallow signals is to accurately locate the signal range to be processed by combining the main characteristics of "early timing, high intensity, and matching with the antenna coupling template" with the depth threshold calculated by wave velocity and knowledge of the field environment. This division can avoid overprocessing of deep signals (preventing the introduction of new noise) and efficiently remove coupling interference from shallow layers, ultimately improving the recognition accuracy of the target signal.

[0033] Then, the computing device can normalize the GPR signal.

[0034] Because the output signal amplitudes of different ground-penetrating radar devices (such as antennas with different center frequencies and acquisition systems with different sampling rates) vary, standardization processing is necessary to ensure data consistency. ; Where X' is the decoupled signal. This is the signal average, used to eliminate DC offset. The standard deviation of the signal is used to standardize the dynamic range of the signal. The first GPR signal after normalization has a distribution that approximates a Gaussian random process with zero mean and unit variance.

[0035] Step 103: The computing device calculates the global characteristics of the first GPR signal by improving the Transformer network.

[0036] The Transformer network is a deep learning model based on the self-attention mechanism. It consists of two parts: an encoder and a decoder, both of which use "attention layers" and "feedforward neural network layers" as core components. The encoder is responsible for converting the input sequence into a "feature representation" containing contextual information, while the decoder is responsible for converting the feature representation output by the encoder into the target sequence.

[0037] This application's embodiments utilize a multi-scale Transformer architecture, combining local and global attention mechanisms to rapidly suppress random surface noise. By incorporating hyperbolic attenuation location coding that simulates the propagation and attenuation characteristics of electromagnetic signals in underground media, long-range stratum reflection features can be accurately extracted.

[0038] The global feature is calculated by the improved Transformer network through local attention and hyperbolic decay position encoding, and the local attention is calculated by the improved Transformer network based on the first GPR signal.

[0039] Preferably, the computing device can extract a short-time feature matrix from the first GPR signal based on a preset sliding window; calculate the local attention based on the short-time feature matrix; perform attention enhancement on the short-time feature matrix based on the local attention; and superimpose the attention-enhanced short-time feature matrix and the hyperbolic decay position code to obtain the global feature.

[0040] The computing device can use a sliding window mechanism (L=5 to 15 sampling points) to extract short-term features, and then calculate local attention based on the matrix of these short-term features.

[0041] ; in, Given the short-time feature matrix as input, Let be the learnable projection matrix, and d be the feature dimension.

[0042] Then, in the deep Transformer global attention module, hyperbolic decay position encoding is introduced: ; Where α is the attenuation coefficient, simulating the attenuation characteristics of electromagnetic waves in a typical medium, f is the frequency modulation parameter, matching the wavelength scale of the GPR main frequency, and d is the propagation distance.

[0043] For example, the improved Transformer network can directly superimpose the attention-enhanced short-time feature matrix and the hyperbolic decay position encoding to obtain global features that simultaneously indicate the time-domain and frequency-domain features of the first GPR signal.

[0044] Traditional denoising methods struggle to accurately resolve GPR signals from superimposed reflections across multiple media. Fixed-scale denoising strategies are ill-suited to simultaneously address shallow high-frequency noise and deep low-frequency interference, leading to the loss of deep target features or phase distortion. This application's embodiments utilize a multi-scale Transformer architecture, combining local and global attention mechanisms to rapidly suppress surface random noise and accurately extract long-range stratigraphic reflection features through deep global attention, achieving differentiated extraction and reconstruction of reflection features at different depths.

[0045] Step 104: The computing device reconstructs the second GPR signal based on the global features using the improved Transformer network.

[0046] After extracting the global features of the first GPR signal, the improved Transformer network can reconstruct the signal based on these global features to obtain the denoised second GPR signal.

[0047] To enhance the model's ability to perceive differences in medium properties, prior physical information such as relative permittivity, conductivity, and density is further embedded into the feature sequence to form physically enhanced features.

[0048] Preferably, before reconstructing the second GPR signal based on the global features, the computing device can obtain the relative permittivity, conductivity, and density of the target corresponding to the original GPR signal; construct the physical constraint code of the target based on the relative permittivity, conductivity, and density; superimpose the physical constraint code and the global features to obtain the enhanced global features; and then reconstruct the second GPR signal based on the enhanced global features.

[0049] The target area is the spatial range corresponding to the original GPR signal, which can specifically refer to the geological or engineering area in the strata where certain phenomena may need to be detected.

[0050] The computing device can obtain the relative permittivity, conductivity, and density from the database of the relevant local agency in the target area, or the operators can collect and detect these data before the detection and preset them in the computing device or a memory connected to the computing device.

[0051] The computing device can then fuse physical constraints, encode this data, and enhance the global features of the input: ; in, As a global feature, The relative permittivity, For electrical conductivity, Let E(·) be the density, and let E(·) be the embedding function implemented by a 3-layer Multi-Layer Perceptron (MLP).

[0052] The performance of fixed-structure denoising models deteriorates sharply in scenarios with abrupt changes in dielectric parameters (such as the permafrost-aquifer interface). The embodiments of this application significantly improve engineering generalization ability by dynamically adjusting the network depth and attention range through online dielectric parameter estimation.

[0053] Preferably, the computing device can reconstruct the second GPR signal using the improved Transformer network based on the global features, reflection guidance mask, and time-frequency joint constraint loss function; wherein, the reflection guidance mask is used to indicate that the attention of the improved Transformer network is focused on the location where geological reflections may occur, and the time-frequency joint constraint loss function is used to optimize the signal reconstruction effect at both the time and frequency domain levels.

[0054] The computing device can construct a hyperbolic reflection-guided mask to correct the attention in the decoding process of the improved Transformer network regarding geological reflection guidance. ; in, The initial amplitude, The medium absorption coefficient, For direct wave cutoff time, It is an indicator function.

[0055] Then, the computing device can optimize and train the model using the time-frequency joint loss function of the multi-physical quantity joint optimization objective. The expression of this time-frequency joint loss function is as follows: ; Where S(·) is the short-time Fourier transform, and W(·) is the smoothed pseudo-Wigner-Ville distribution. Here, N is the tuning parameter, and N is the total number of samples. Represents the i-th original sample. Let i represent the i-th reconstructed sample. Describing the Frobenius norm, This indicates the method of spectral distance measurement.

[0056] For example, The values ​​are 0.5 and 0.3 respectively.

[0057] Traditional deep learning methods (such as U-Net) rely solely on data-driven approaches, which can easily disrupt the time-frequency characteristics of electromagnetic wave reflection, leading to hyperbolic shape distortion or phase shift. This application's embodiments introduce hyperbolic reflection guidance masks to guide attention and a joint time-frequency constraint loss, forcing the network output to conform to the wave equation.

[0058] Building upon step 103, a propagation path P(d) and a hyperbolic reflection guidance mask M(t) are introduced to regulate the attention of the Transformer network's encoding and decoding process, enabling the model to focus on the possible locations and scales of geological reflections. Finally, the fused multi-scale features are reconstructed within a multi-layer attention network, and the signal reconstruction effect is optimized at both the time and frequency domains using a joint time-frequency loss function. This process maintains a progressive structure from local perturbation modeling and global attenuation modeling to medium-guided modeling, effectively improving the model's ability to identify GPR noise and weak reflective targets.

[0059] Step 105: The computing device smooths the second GPR signal and / or removes island noise to obtain the target GPR signal.

[0060] Although deep denoising networks can effectively suppress most noise, high-frequency residual noise may still exist in the low-frequency region of the signal (such as deep weak reflections).

[0061] Preferably, the computing device uses a time-varying Gaussian filter to smooth the second GPR signal.

[0062] To enhance signal continuity without compromising effective quality reflection characteristics, this method employs time-varying Gaussian filtering for adaptive smoothing. ; in: The time-domain waveform of the input second GPR signal. It is a time-varying Gaussian kernel function. Time-varying smoothing parameter (unit: number of sampling points).

[0063] This approach not only meets the high-precision requirements of shallow layers, avoiding disruption of short-term reflection characteristics (such as steel mesh and shallow pipelines), but also optimizes reliability for deeper layers. Deep signals are typically affected by medium absorption, exhibiting strong continuity but low signal-to-noise ratio; appropriate smoothing can improve the visual coherence of the in-phase axis.

[0064] Preferably, the computing device can use dynamic thresholding based on sliding statistics to remove island noise in the second GPR signal.

[0065] Among them, the noise in the second GPR signal often exhibits non-stationary characteristics, that is, the noise level is inconsistent at different depths. Island noise refers to noise components that appear in isolation in the signal and have significantly different characteristics from the surrounding signals. They are usually manifested as spikes or glitches with small amplitude and short duration.

[0066] Preferably, the computing device can employ dynamic thresholding based on sliding statistics to ensure that only valid reflected signal components are retained: ; Among them, dynamic noise level estimation Statistics based on window length during window movement Sampling points) calculation: ; in, Multiples of the threshold It is a local mean. Binary mask (1 = valid signal, 0 = noise), x(t) is the input second GPR signal.

[0067] By noise tagging, all Points that are not valid are considered noise and replaced with interpolated values ​​from nearby valid points. If the length of a continuous noise segment is less than a preset length threshold, it is determined to be spurious noise and the original signal is preserved.

[0068] In this embodiment, by improving the Transformer architecture network to calculate local attention and combining it with hyperbolic decay position encoding to calculate global features, the reflection features of long-range strata can be accurately extracted, and the differentiated extraction and reconstruction of reflection features at different depths can be achieved.

[0069] The method provided in the embodiments of this application has been described above. The system provided in the embodiments of this application will be described below.

[0070] Please see Figure 2 , Figure 2A schematic diagram of a ground-penetrating radar signal denoising system based on an improved Transformer, provided as an embodiment of this application, is shown below. Figure 2 As shown, the system 20 includes: Acquisition module 201 is used to acquire raw ground-penetrating radar GPR signals; The first processing module 202 is used to perform noise reduction preprocessing on the original GPR signal to obtain the first GPR signal; The second processing module 203 is used to calculate the global features of the first GPR signal through an improved Transformer network; wherein the global features are calculated by the improved Transformer network through local attention and hyperbolic attenuation position encoding, and the local attention is calculated by the improved Transformer network based on the first GPR signal; The second processing module 203 is also used to reconstruct the second GPR signal based on the global features through the improved Transformer network; The third processing module 204 is used to smooth the second GPR signal and / or remove island noise to obtain the target GPR signal.

[0071] Preferably, the second processing module 203 is specifically used to extract a short-time feature matrix from the first GPR signal based on a preset sliding window; calculate the local attention based on the short-time feature matrix; perform attention enhancement on the short-time feature matrix based on the local attention; and superimpose the attention-enhanced short-time feature matrix and the hyperbolic decay position code to obtain the global feature.

[0072] Preferably, the acquisition module 201 is further used to acquire the relative permittivity, conductivity, and density of the target corresponding to the original GPR signal; the system 20 also includes a physical constraint module 205, used to construct a physical constraint code for the target based on the relative permittivity, conductivity, and density; superimpose the physical constraint code and the global feature to obtain the enhanced global feature; the second processing module 203 is specifically used to reconstruct the second GPR signal based on the enhanced global feature.

[0073] Preferably, the second processing module 203 is specifically used to reconstruct the second GPR signal through the improved Transformer network based on the global features, reflection guidance mask, and time-frequency joint constraint loss function; wherein, the reflection guidance mask is used to indicate that the attention of the improved Transformer network is focused on the location where geological reflection may occur, and the time-frequency joint constraint loss function is used to optimize the signal reconstruction effect at both the time domain and frequency domain levels.

[0074] Preferably, the first processing module 202 is specifically used to perform time-varying gain on the original GPR signal based on a logarithmic function; remove the antenna coupling interference component in the original GPR signal after time-varying gain based on a preset antenna coupling response template; and normalize the original GPR signal after removing the antenna coupling interference component to obtain the first GPR signal.

[0075] Preferably, the third processing module 204 is specifically used to smooth the second GPR signal using a time-varying Gaussian filter.

[0076] Preferably, the third processing module 204 is specifically used to remove island noise in the second GPR signal by employing dynamic threshold segmentation based on sliding statistics.

[0077] The ground-penetrating radar signal denoising system based on the improved Transformer provided in this application can be understood by referring to the relevant content in the foregoing method embodiment section, and will not be repeated here.

[0078] like Figure 3 As shown, Figure 3 This is a schematic diagram of a possible logical structure of a computing device provided in an embodiment of this application. The computing device 30 includes a processor 301, a communication interface 302, a memory 303, and a bus 304. The processor 301, the communication interface 302, and the memory 303 are interconnected via the bus 304. In an embodiment of this application, the processor 301 is used to control and manage the operation of the computing device 30. For example, the processor 301 is used to execute... Figure 1 The steps in the embodiments and / or other processes used in the techniques described herein. Communication interface 302 is used to support communication by computing device 30. Memory 303 is used to store program code and data of computing device 30.

[0079] The processor 301 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, etc. The bus 304 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0080] In another embodiment of this application, a computer-readable storage medium is also provided, the computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the above-described... Figure 1 The method described in the embodiments.

[0081] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0082] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0083] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0084] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for denoising ground-penetrating radar signals based on an improved Transformer, characterized in that, The method includes: Acquire raw ground-penetrating radar (GPR) signals; The original GPR signal is subjected to denoising preprocessing to obtain the first GPR signal; The global features of the first GPR signal are calculated by an improved Transformer network; wherein the global features are calculated by the improved Transformer network through local attention and hyperbolic attenuation position encoding, and the local attention is calculated by the improved Transformer network based on the first GPR signal; The improved Transformer network is used to reconstruct the second GPR signal based on the global features. The second GPR signal is smoothed and / or island noise is removed to obtain the target GPR signal.

2. The method according to claim 1, characterized in that, The step of calculating the global features of the first GPR signal by improving the Transformer network includes: Based on a preset sliding window, a short-time feature matrix is ​​extracted from the first GPR signal; The local attention is calculated based on the short-time feature matrix; The short-term feature matrix is ​​enhanced with attention based on the local attention; The global features are obtained by superimposing the attention-enhanced short-term feature matrix and the hyperbolic decay position code.

3. The method according to claim 1, characterized in that, Before reconstructing the second GPR signal based on the global features, the method further includes: Obtain the relative permittivity, conductivity, and density of the target corresponding to the original GPR signal; Based on the relative permittivity, the conductivity, and the density, a physical constraint code for the detection target is constructed. By superimposing the physical constraint encoding and the global features, the enhanced global features are obtained; The process of reconstructing the second GPR signal based on the global features includes: The second GPR signal is obtained by reconstructing based on the enhanced global features.

4. The method according to claim 1, characterized in that, The process of reconstructing the second GPR signal based on the global features using the improved Transformer network includes: The improved Transformer network is used to reconstruct the second GPR signal based on the global features, reflection guidance mask, and time-frequency joint constraint loss function. The reflection guidance mask is used to instruct the improved Transformer network to focus its attention on the locations where geological reflections may occur, and the time-frequency joint constraint loss function is used to optimize the signal reconstruction effect at both the time and frequency domains.

5. The method according to claim 1, characterized in that, The step of performing denoising preprocessing on the original GPR signal to obtain the first GPR signal includes: The original GPR signal is subjected to a time-varying gain based on a logarithmic function; Based on a preset antenna coupling response template, the antenna coupling interference component in the original GPR signal after time-varying gain is removed; The original GPR signal, after removing the antenna coupling interference component, is normalized to obtain the first GPR signal.

6. The method according to claim 1, characterized in that, The smoothing and / or island noise removal of the second GPR signal includes: The second GPR signal is smoothed by using a time-varying Gaussian filter.

7. The method according to claim 1, characterized in that, The smoothing and / or island noise removal of the second GPR signal includes: Dynamic threshold segmentation based on sliding statistics is used to remove island noise from the second GPR signal.

8. A ground-penetrating radar signal denoising system based on an improved Transformer, characterized in that, The system, applied to the method of any one of claims 1-7, comprises: The acquisition module is used to acquire the raw ground-penetrating radar GPR signal; The first processing module is used to perform noise reduction preprocessing on the original GPR signal to obtain the first GPR signal; The second processing module is used to calculate the global features of the first GPR signal through an improved Transformer network; wherein the global features are calculated by the improved Transformer network through local attention and hyperbolic attenuation position encoding, and the local attention is calculated by the improved Transformer network based on the first GPR signal; The second processing module is further configured to reconstruct the second GPR signal based on the global features using the improved Transformer network; The third processing module is used to smooth the second GPR signal and / or remove island noise to obtain the target GPR signal.

9. A computing device, characterized in that, include: Memory, used to store programs; A processor for loading the program to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method of any one of claims 1-7.

Images