Ground penetrating radar data multi-frequency fusion method, device, equipment, medium and product

Abstract

The application discloses a ground penetrating radar data multi-frequency fusion method, device, equipment, medium and product, relates to the field of ground penetrating radar data processing, and comprises the following steps: acquiring multi-frequency ground penetrating radar data; the multi-frequency ground penetrating radar data comprises at least two ground penetrating radar data with different frequency characteristics; the multi-frequency ground penetrating radar data is preprocessed to obtain preprocessed multi-frequency data; the preprocessed multi-frequency data is mapped across frequencies by using a Cycle-GAN network to obtain pseudo-high-frequency ground penetrating radar data; and the pseudo-high-frequency ground penetrating radar data and actual high-frequency data are adaptively fused by using a Transformer fusion module to obtain a weighted fusion result. The application can improve the quality of multi-frequency data fusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ground penetrating radar data processing, and particularly relates to a ground penetrating radar data multi-frequency fusion method, device, equipment, medium and product. BACKGROUND

[0002] The performance of a ground penetrating radar (GPR) system is limited by the inherent constraint relationship between the antenna frequency and the detection depth and spatial resolution: a high-frequency antenna (such as 2GHz) can achieve millimeter-level shallow layer resolution, but the signal attenuation is serious and it is difficult to penetrate deep media; while a low-frequency antenna (such as 50MHz) has a detection depth of tens of meters, but its resolution cannot meet the fine target identification requirements. The multi-frequency data fusion technology is developed to solve the above-mentioned contradiction, and in the field of GPR multi-frequency data fusion, there is still a problem of low data fusion quality. SUMMARY

[0003] The purpose of the present application is to provide a ground penetrating radar data multi-frequency fusion method, device, equipment, medium and product, which can improve the quality of multi-frequency data fusion.

[0004] To achieve the above-mentioned purpose, the present application provides the following solutions.

[0005] In a first aspect, the present application provides a ground penetrating radar data multi-frequency fusion method, comprising: acquiring multi-frequency ground penetrating radar data; the multi-frequency ground penetrating radar data comprises at least two ground penetrating radar data with different frequency characteristics; pre-processing the multi-frequency ground penetrating radar data to obtain pre-processed multi-frequency data; performing cross-frequency mapping on the pre-processed multi-frequency data using a Cycle-GAN network to obtain pseudo-high-frequency ground penetrating radar data; and performing adaptive fusion on the pseudo-high-frequency ground penetrating radar data and actual high-frequency data using a Transformer fusion module to obtain a weighted fusion result.

[0006] In an embodiment, the pre-processing includes zero-time correction, direct wave removal, time-varying gain processing and amplitude normalization.

[0007] In an embodiment, the loss function of the Cycle-GAN network is a weighted sum of a bidirectional adversarial loss, a cycle consistency loss and a multi-scale structural similarity loss.

[0008] In an embodiment, the loss function of the Transformer fusion module includes a weighted pixel-level error and a structural similarity loss.

[0009] In an embodiment, the calculation process of the loss function of the Transformer fusion module specifically comprises: determining high-frequency component prior weights and spatial gradient prior weights according to the pseudo high-frequency ground penetrating radar data and the actual high-frequency data; determining a joint weight according to the high-frequency component prior weights and the spatial gradient prior weights; and determining a weighted pixel-level error and a structural similarity loss according to the joint weight.

[0010] In an embodiment, the Transformer fusion module comprises a multi-band feature embedding module, a multi-head attention extraction module and a dynamic weight prediction module connected in sequence.

[0011] In a second aspect, the present application provides a ground penetrating radar data multi-frequency fusion device, comprising: an acquisition module configured to acquire multi-frequency ground penetrating radar data; the multi-frequency ground penetrating radar data comprises ground penetrating radar data of at least two different frequency characteristics; a preprocessing module configured to preprocess the multi-frequency ground penetrating radar data to obtain preprocessed multi-frequency data; a cross-frequency mapping module configured to perform cross-frequency mapping on the preprocessed multi-frequency data using a Cycle-GAN network to obtain pseudo high-frequency ground penetrating radar data; and an adaptive fusion module configured to perform adaptive fusion on the pseudo high-frequency ground penetrating radar data and actual high-frequency data using a Transformer fusion module to obtain a weighted fusion result.

[0012] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the ground penetrating radar data multi-frequency fusion method.

[0013] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the ground penetrating radar data multi-frequency fusion method.

[0014] In a fifth aspect, the present application provides a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the ground penetrating radar data multi-frequency fusion method.

[0015] According to the specific embodiments provided by the present application, the following technical effects are disclosed.

[0016] The application provides a ground penetrating radar data multi-frequency fusion method, device, equipment, medium and product. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0018] Figure 1 It is a schematic diagram of the ground penetrating radar data multi-frequency fusion method.

[0019] Figure 2 It is a schematic diagram of the simulation data preprocessing of gprMax.

[0020] Figure 3 It is a simulation data preprocessing result graph.

[0021] Figure 4 It is a structure diagram of the Cycle-GAN network.

[0022] Figure 5 It is a structure diagram of the Transformer fusion module.

[0023] Figure 6 It is a simulation experiment result graph.

[0024] Figure 7 It is a measured data result graph.

[0025] Figure 8 It is a flowchart of the ground penetrating radar data multi-frequency fusion method.

[0026] Figure 9 It is a functional module schematic diagram of the ground penetrating radar data multi-frequency fusion device.

[0027] Figure 10 It is a structure schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0028] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of the present application.

[0029] In the field of GPR multi-frequency data fusion, traditional methods mainly include time domain fusion, frequency domain fusion and time-frequency domain fusion. The time domain fusion method is intuitive and efficient, but it is difficult to accurately quantify the contribution of different frequency components in the time domain. The frequency domain and time-frequency domain methods rely on accurate modeling of antenna characteristic parameters and medium attribute information, and parameter estimation inaccuracy can introduce fusion errors. Moreover, most methods lack in-depth modeling of multi-scale spatial structure and high-level semantic information of images, making it difficult to effectively preserve shallow details and enhance deep target expression.

[0030] In recent years, deep learning technology has shown potential in this field, but existing methods still face challenges. Some methods, such as Cycle-GAN, lack the guidance of shallow real high-frequency data, and may produce artifacts or lose high-frequency details when reconstructing shallow targets. Some methods, such as Transformer, can preserve shallow high-frequency information, but cannot improve the spatial resolution after fusion, making it difficult to meet the needs of advanced applications. Therefore, there is an urgent need for a new GPR multi-frequency data fusion method to address the shortcomings of existing technology.

[0031] To make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0032] In one exemplary embodiment, as shown in Figure 8 a ground penetrating radar data multi-frequency fusion method is provided, comprising the following steps.

[0033] Step 801: Obtain multi-frequency ground penetrating radar data; the multi-frequency ground penetrating radar data includes at least two ground penetrating radar data with different frequency characteristics.

[0034] Step 802: Preprocess the multi-frequency ground penetrating radar data to obtain preprocessed multi-frequency data.

[0035] Step 803: Perform cross-frequency mapping on the preprocessed multi-frequency data using a Cycle-GAN network to obtain pseudo-high-frequency ground penetrating radar data.

[0036] Step 804: Perform adaptive fusion on the pseudo-high-frequency ground penetrating radar data and actual high-frequency data using a Transformer fusion module to obtain a weighted fusion result. In practical applications, the actual high-frequency data is a real high-frequency image.

[0037] The Cycle-GAN network and the Transformer fusion module are combined to realize effective conversion of low frequency to high frequency and adaptive fusion of high frequency, which can retain shallow texture and edge information and enhance the expression ability of deep structure, thereby improving the fusion quality of data.

[0038] In one exemplary embodiment, the pre-processing includes zero-time correction, direct wave removal, time-varying gain processing, and amplitude normalization.

[0039] In one exemplary embodiment, the loss function of the Cycle-GAN network is a weighted sum of bidirectional adversarial loss, cycle consistency loss, and multi-scale structural similarity loss.

[0040] In one exemplary embodiment, the loss function of the Transformer fusion module includes a weighted pixel-level error and a structural similarity loss.

[0041] In actual application, the calculation process of the loss function of the Transformer fusion module specifically includes: determining high-frequency component prior weight and spatial gradient prior weight according to pseudo-high-frequency ground penetrating radar data and actual high-frequency data; determining a joint weight according to the high-frequency component prior weight and the spatial gradient prior weight; determining a weighted pixel-level error and a structural similarity loss according to the joint weight.

[0042] In actual application, the Transformer fusion module includes a multi-band feature embedding module, a multi-head attention extraction module, and a dynamic weight prediction module connected in sequence.

[0043] As shown in Figure 1 The core idea can be summarized as: how to synergistically use the deep penetration advantage of low-frequency data and the shallow high-resolution capability of high-frequency data to realize high-quality data fusion in the full depth domain.

[0044] The present application discloses a multi-frequency fusion method based on multi-scene ground penetrating radar data.

[0045] In one exemplary embodiment, as shown in Figure 1 A multi-frequency fusion method for ground penetrating radar data is provided to improve GPR image quality and information fidelity, and to meet the detection needs in complex actual environments. The method includes the following steps.

[0046] Step 1: Obtain GPR multi-frequency data; wherein the multi-frequency data need to include at least two ground penetrating radar data with different frequency characteristics.

[0047] Step 2: Multi-frequency ground penetrating radar data pre-processing, and constructing a multi-frequency fusion data set based on the pre-processed ground penetrating radar data.

[0048] Step 3: Realize the cross-frequency mapping from low-frequency image to high-frequency domain based on Cycle-GAN. In the training process, the input is the preprocessed high and low frequency ground penetrating radar image obtained in step 2, and the output is the pseudo high frequency ground penetrating radar image. In practical application, the pseudo high frequency ground penetrating radar image is output by using the low frequency ground penetrating radar image.

[0049] Step 4: Use the Transformer fusion module to adaptively fuse the high-frequency features of the pseudo high-frequency image generated by Cycle-GAN and the real high-frequency image, and finally output the weighted fusion result of the measured data and the pseudo high-frequency data. The "real high-frequency image" referred to in the present application is derived from the high-frequency component in the multi-frequency data of ground penetrating radar (GPR). The original radar signal is stored in the form of a matrix, representing the electromagnetic properties of the underground medium. Through digital signal processing (such as gain, filtering, migration, etc.) and visualization technology, this data matrix can be converted into a grayscale image or a pseudo-color image for training by a deep learning network. For ease of description, the original matrix without visualization processing is referred to as "data" in the present application, and the visual representation generated after processing is referred to as "image".

[0050] Multi-frequency ground penetrating radar data includes: ground penetrating radar simulation data obtained by using gprMax simulation software and professional ground penetrating radar measured data, and at least one of the data is required. The measured data is collected by IDS RIS-K2. The frequency includes 400MHz, 900MHz and 1600MHz. Multi-frequency ground penetrating radar data: high frequency data and low frequency data can be paired or unpaired, and the acquisition approach can include gprMax software simulation and field measurement.

[0051] Preprocessing includes zero-time correction, direct wave removal, time-varying gain processing and amplitude normalization of the original data; for simulation data, Gaussian white noise is added before data gain to simulate the noise interference in real world GPR data; for measured data, time zero correction, automatic gain control and amplitude normalization processing are performed to ensure the comparability and consistency of each data set in subsequent fusion analysis.

[0052] Cycle-GAN network contains two sets of generator-discriminator pairs, generator and respectively realize the bidirectional conversion of low frequency domain (A) to high frequency domain (B). The converted data is compared with the real data through discriminators and to improve the fidelity of the generated image, and then mapped back to the original domain through inverse mapping, and the cycle consistency loss is calculated with the initial data to ensure the physical interpretability of the generated result. The generator adopts the U-net structure, which consists of an encoder, a converter and a decoder.

[0053] The encoder is used to extract the features of the input image. It consists of three convolutional layers that progressively compress the image size and increase the number of channels to extract higher-level features. This layer has 64 filters, each with a size of 7x7 and a stride of 1, using a ReLU activation function. It performs preliminary feature extraction on the input image, typically capturing basic features such as edges and textures. Using a larger 7x7 convolution kernel allows capturing a larger range of contextual information and basic texture features in the image at the initial stage. A stride of 1 means that the spatial dimensions of the image (KxK) are not changed, mainly mapping the input image from 3 channels to 64 feature channels for preliminary feature extraction. The ReLU activation function introduces nonlinearity, allowing the network to learn more complex patterns. The second layer has 128 filters, each with a size of 3x3 and a stride of 2, using a ReLU activation function. By using a stride of 2 convolution, the image size is compressed to 1 / 2 of the original size, which helps to expand the receptive field of subsequent layers, allowing the network to focus on more macroscopic structures. This step effectively reduces the computational load while extracting features. The third layer has 256 filters, each with a size of 3x3 and a stride of 2, using a ReLU activation function. The image size is further compressed to 1 / 4 of the original size, while the number of channels increases to 256, used to extract more abstract and complex features. By this point, the encoder has compressed the original image into a tensor with a smaller size but a high feature dimension, which contains the core content and structural information of the original low-frequency image.

[0054] The converter is the core part of the generator, responsible for converting the features extracted by the encoder into features of the target domain. 8 residual blocks: each module contains 2 3x3 convolution layers (256 convolution kernels, stride s=1), and a "skip connection" is established between the input and output of the module. Use ReLU activation function. These modules use skip connections to directly add the input to the output, which can effectively alleviate the problem of gradient disappearance, allowing the network to be trained deeper and thus learn more complex mapping relationships. They are responsible for deep transformation of the features extracted by the encoder, converting the features of the low-frequency image into the features of the high-frequency image. It adjusts and reorganizes the features without changing the size of the feature map, achieving "style" conversion.

[0055] The function of the decoder is the opposite of the encoder. It receives the high-dimensional features processed by the transformer and gradually recovers the spatial resolution and details of the image through a series of deconvolution (or transpose convolution) operations, finally generating the target high-frequency image. The first layer is a deconvolution layer with 128 filters, each filter size is 3x3, step is 2, and uses ReLU activation function. It restores the image size to 1 / 2 of the original size and reduces the channel number to 128. The second layer is also a deconvolution layer with 64 filters, each filter size is 3x3, step is 2, and uses ReLU activation function. It restores the image size to the original size. The third layer is the last layer with 3 filters, each filter size is 7x7, step is 1, and uses ReLU activation function. It restores the channel number to 3 (usually corresponding to RGB images) and finally outputs the synthesized high-frequency image. Using a large convolution kernel of 7x7 helps to integrate larger range of features and generate smooth and natural images.

[0056] The design goal of the discriminator is to determine whether an input image is "real" (from the real data set) or "fake" (synthesized by the generator). The structure adopted is a PatchGAN discriminator. The traditional discriminator outputs a single "true / false" probability value for the entire image. The PatchGAN discriminator outputs an N x N feature map, where each element (or "pixel") corresponds to a local region (Patch) in the input image and determines the authenticity of the region. The advantage of this design is that it can more effectively determine whether the local details and textures of the image are realistic, which is very important for GPR images which are rich in texture information, and the parameter quantity is relatively small and the running speed is faster.

[0057] Input layer: receives an image, which can be a real high-frequency GPR image or a fake high-frequency image generated by the generator G_AB. The first layer of convolution has 64 4x4 convolution kernels with a stride of s=2 and uses a LeakyReLU activation function. The input image is subjected to preliminary feature extraction and down-sampling. The combination of 4x4 convolution kernels and a stride of 2 is common in discriminators. LeakyReLU is an improved ReLU that allows a small non-zero slope when the input is negative, which helps prevent neuron "death" and enables better gradient flow during training, enhancing the stability of the discriminator training. The second layer of convolution has 128 4x4 convolution kernels with a stride of s=2 and uses LeakyReLU. Further extraction of deeper-level features is performed, and the feature map size is halved. The number of channels increases to capture more complex patterns. The third layer of convolution has 256 4x4 convolution kernels with a stride of s=2 and uses LeakyReLU. The feature extraction process continues to deepen, and the feature map size is again halved. The fourth layer of convolution has 512 4x4 convolution kernels with a stride of s=2 and uses LeakyReLU. This is the deepest layer of feature extraction, with the most feature channels (512), capable of capturing very abstract and complex discriminative features. The fifth layer (output layer): has 1 4x4 convolution kernel with a stride of s=1. This is the final discriminative layer. It compresses the 512 high-dimensional feature maps extracted by the previous layer into a single-channel feature map (i.e., N x N x 1). Each value of this output map represents the "authenticity" score of the discriminator for a corresponding local region (Patch) of the input image. During training, the network aims to make the value of the real image on this output map approach 1, while the value of the fake image approaches 0. This output map is called the Patch-Guided Discriminator.

[0058] The loss function of this module is composed of adversarial loss, cycle consistency loss, and multi-scale structural similarity loss. The final total loss function is the weighted sum of the three, and the hyperparameters are determined by heuristic search to balance model complexity and fitting performance. The quality and physical interpretability of the generated image are optimized through a specific loss function.

[0059] In one embodiment, Cycle-GAN is used to realize the cross-frequency mapping of low-frequency images to high-frequency domain, and the input size of the model is unified to 256x256 pixel values. The loss function of this module is composed of three parts: adversarial loss, cycle consistency loss, and multi-scale structural similarity loss. The adversarial loss is optimized by the generator and the discriminator together. The generator aims to minimize the probability of the generated image being recognized by the discriminator, and the discriminator aims to maximize the discrimination ability of the real and generated images. The bidirectional adversarial loss ( ) is defined as follows:

[0060] .

[0061] where the formula represents the bidirectional adversarial loss, respectively, the generator loss and discriminator loss of the low-frequency domain to the high-frequency domain, and the generator loss and discriminator loss of the high-frequency domain to the low-frequency domain. a, b represent the high, low frequency dataset samples, and respectively represent the generation process from the low-frequency domain to the high-frequency domain and the high-frequency domain to the low-frequency domain, and The discriminator corresponding to the domain is shown, and a~A and b~B represent the sampling from the low-frequency domain and the high-frequency domain. is the adversarial loss function.

[0062] Cycle consistency loss Cycle-GAN network is the core mechanism, which ensures the generalization ability of the generator and reduces the dependence on paired data. The specific formula is as follows:

[0063] .

[0064] Multi-scale structural similarity loss From the three aspects of data details, contrast and brightness, the structure of the generated data is constrained to suppress artifacts, which has been proved to have the effect of deblurring. The specific calculation formula is as follows:

[0065] .

[0066] .

[0067] respectively represent the mean and variance and covariance of images x and y, a, b represent the high, low frequency dataset samples, and respectively represent the generation process from the low-frequency domain to the high-frequency domain and the high-frequency domain to the low-frequency domain. represent the calculation and the multi-scale structural loss of b, represent the calculation and the weighted multi-scale structural loss of a. Wherein, MS-SSIM uses five different scales of SSIM calculation, and the weight decays according to the depth: .

[0068] Finally, the total loss function is the weighted sum of the above three parts:

[0069] .

[0070] where the hyperparameters λ and γ are determined by a heuristic search to balance the model complexity and the fitting performance. Experiments show that when λ = 0.1 and γ = 0.001, the model achieves the optimal balance between training stability and objective function convergence. During the optimization process, all experiments use an initial learning rate of 0.002, which remains constant for the first 100 epochs and then decays linearly to zero for the next 100 epochs to accelerate convergence and avoid overfitting.

[0071] The Transformer fusion module in step 4 includes a multi-band feature embedding module, a multi-head attention extraction module, and a dynamic weight prediction module. The multi-band feature embedding module inputs the original high-frequency image and the pseudo high-frequency image into their respective embedding layers, maps them to a unified semantic space, generates feature vector representations, and concatenates them to obtain joint feature representations; the multi-head attention extraction module uses a multi-head self-attention mechanism to model the long-range dependencies between different time steps, capture response patterns, and obtain comprehensive feature representations containing multi-dimensional and multi-level context information; the dynamic weight prediction module generates adaptive fusion weights through a feedforward layer and a linear layer.

[0072] Step 4 is used to design a time-frequency optimized loss function, including high-frequency component prior weight extraction, spatial gradient prior weight extraction, joint weight calculation, and construction of the final loss function. Perform short-time Fourier transform on the input signal to obtain time-frequency representation, define the high-frequency prior weight as the energy within the preset high-frequency band; use a second-order difference operator to detect the sharp changes in the spatial gradient, generating a spatial gradient prior weight; linearly combine the high-frequency prior and the spatial gradient weight to generate a space-frequency joint weight; the final loss function is composed of weighted pixel-level error and structural similarity loss, guiding the model to focus on areas that need accurate reconstruction, improving the fidelity of the generated image in high-frequency details and structural boundaries, and accelerating the convergence speed.

[0073] In one embodiment, the Transformer fusion module in step 4 above consists of 3 key components: 1) Multi-band feature embedding module: input the original high-frequency image and the pseudo high-frequency image converted by Cycle-GAN into their respective embedding layers, map them to a unified semantic space, and generate feature vector representations.

[0074] .

[0075] .

[0076] where , , and are learnable weight matrices, and the joint feature representation , the original GPR data is converted into a feature sequence containing both data content information and spatial position information.

[0077] 2) Multi-head attention extraction module: a multi-head self-attention mechanism is used to model the long-range dependencies between different time steps, capture the response patterns from the measured data and the generated data at multiple time scales, and adaptively identify the areas where the real signal needs to be preserved and the areas where enhanced high-frequency details are suitable. The attention mechanism is calculated based on the scaled dot-product attention. In order to calculate the self-attention, the feature sequence E is generated through three independent learnable linear projection matrices , , , respectively, to generate query (Query, Q), key (Key, K) and value (Value, V) matrices:

[0078] .

[0079] .

[0080] .

[0081] On this basis, the multi-head attention mechanism calculates the attention in parallel by performing h independent linear projections on Q, K, V, and the process is as follows.

[0082] .

[0083] .

[0084] MultiHead(Q, K, V) .

[0085] where Q (query), K (key), and V (value) come from the feature sequence output by the previous stage. is the learnable projection matrix of the i-th head, indicates the output of the i-th attention head, and h indicates the number of attention heads. The outputs of all heads are concatenated through the Concat operation, and the final projection matrix is used for fusion to obtain a comprehensive feature representation MultiHead(Q, K, V) containing multi-dimensional and multi-level context information, denoted as H. Each element in H encodes the fusion decision basis of the original high-frequency signal and the pseudo high-frequency signal under global association, laying a solid foundation for subsequent adaptive fusion weight prediction.

[0086] 3) Dynamic weight prediction module: adaptive fusion weights are generated through a feedforward layer and a linear layer .

[0087] .

[0088] where is the hidden state of both real high-frequency data and pseudo high-frequency data, is the residual, is the Sigmoid function, and the final output result is the weighted fusion of the measured data and the pseudo high-frequency data:

[0089] .

[0090] where, ⊙ represents the Hadamard product, which ensures the consistency of the fusion process in time and amplitude through element-by-element multiplication, represents the original high-frequency image, represents the pseudo high-frequency image.

[0091] In one embodiment, the loss function of the above-mentioned step 4 fusion module specifically includes the following parts.

[0092] 1) High-frequency component prior weight Extraction: The high-frequency component in the ground penetrating radar image usually corresponds to the scattering signals of fine bedding, texture and small abnormal bodies, and is the key to improving image resolution. Short-time Fourier transform (STFT) is performed on the input signal (A-scan) to obtain its time-frequency representation .

[0093] .

[0094] where, is the ith A-scan, is the window function, which is selected as the Hamming window to suppress spectral leakage. Then, the high-frequency prior weight is defined as the energy in the preset high-frequency band . is the lower limit of the effective frequency band, is the upper limit of the effective frequency band, represents the actual frequency of the data. Taking 900MHz data as an example, and are taken as 800 and 1000 respectively, which enables the model to focus on those spatial and temporal positions that should contain rich high-frequency information in physics.

[0095] .

[0096] 2) Spatial gradient prior weight Extraction: The reflection of targets such as stratigraphic interfaces, pipelines or root systems in GPR images usually appears as a sudden change in signal. To highlight these important structural boundaries, a second-order difference operator (i.e., a one-dimensional Laplacian) is used to detect the sharp changes in spatial gradients. This operator generates spatial gradient prior weights by convolving with the image .

[0097] .

[0098] 3) Joint weight computation and final loss function: The high-frequency prior and the spatial gradient weights are linearly combined to generate spatial-frequency joint weights .

[0099] .

[0100] Final loss function is composed of weighted pixel-wise error and structural similarity loss .

[0101] .

[0102] .

[0103] .

[0104] .

[0105] is the structural similarity loss, and denotes element-wise multiplication. With this design, this loss function can intelligently direct the model’s “attention” to the areas that need the most accurate reconstruction. This frequency- and gradient-prior-based targeted optimization strategy not only significantly improves the fidelity of the generated images in high-frequency details and structural boundaries but also provides a more explicit optimization direction for the model, effectively accelerating the convergence speed.

[0106] This application acquires ground-penetrating radar (GPR) data, which needs to include high- and low-frequency data of different frequencies. The acquisition method can be simulated using gprMax software or collected in the field by a professional GPR system. The acquired data is preprocessed, and a multi-frequency GPR dataset is constructed based on the preprocessed data. The multi-frequency GPR data fusion features are dominated by high-frequency information in shallow layers and by low-frequency information in deeper layers. Multiple training datasets from different frequencies and scenarios are input into a Cycle-GAN-based cross-frequency mapping module for frequency domain transformation and time domain alignment. A Transformer network is used to fuse the synthesized pseudo-high-frequency data with the real high-frequency data to obtain the final fusion result. The trained model is then applied to the fusion of GPR data in different scenarios. By combining Cycle-GAN with Transformer, we achieve efficient conversion of low-frequency images to the high-frequency domain and adaptive fusion of high- and low-frequency data. This preserves shallow texture and edge information while enhancing the expressive power of deep structures. The designed time-frequency optimized loss function improves the fidelity of the generated images in high-frequency details and structural boundaries. Validated on multi-scene datasets, it exhibits good robustness and universality, providing an effective solution for multi-frequency GPR image fusion in complex media environments. It can be widely applied in fields such as geological exploration and engineering inspection.

[0107] In one embodiment, such as Figure 2 As shown, where, Figure 2 (a) Figure 2 (b) and Figure 2 (c) in the figure represents the original simulation data for 400MHz, 900MHz, and 1600MHz, respectively. Figure 2 (d) Figure 2 (e) and Figure 2 (f) represents the preprocessed data at 400MHz, 900MHz, and 1600MHz, respectively. Figure 2 (g) in Figure 2 (h) and Figure 3 In the diagram (i), the results are shown after adding Gaussian noise that increases with depth. The first row shows the data without preprocessing. Due to the inherent attenuation of electromagnetic waves propagating in the medium, the reflected signal energy of deep targets is very weak and almost invisible in the image. The second row shows the results after data preprocessing. It can be seen that the direct wave is effectively suppressed, and through gain compensation, the hyperbolic reflection characteristics of deep targets are significantly enhanced, and the signal-to-noise ratio is greatly improved. The third row shows the results after adding Gaussian white noise. This processing method significantly amplifies the background noise in the deep layer (below 15 ns) by the gain function, resulting in a significantly lower signal-to-noise ratio (SNR) in the deep layers of the image compared to the shallow layers. This is highly consistent with the actual distribution characteristics of the data collected on-site.

[0108] In one embodiment, such as Figure 3 As shown, Figure 3 (a) and Figure 3 (b) shows the 1600MHz and 900MHz data collected in the Mu Us Desert. After data preprocessing, the 1600MHz high-frequency data displayed extremely high resolution, capable of finely depicting the subtle layering structure of the near-surface (approximately 0-10 ns) soil medium. However, this came at the cost of severely insufficient penetration depth, with almost no effective signal at deeper levels. Figure 4 Although the 900 MHz data in (b) is not as good as 1600 MHz in terms of shallow detail resolution, its detection depth is significantly increased, clearly revealing the stratigraphic structure at a deeper level (about 10-20 ns).

[0109] In one embodiment, such as Figure 5 The diagram illustrates the basic structure of the Cycle-GAN cross-frequency mapping module. It includes a generator and a discriminator, optimizing the quality and physical interpretability of the generated images through a specific loss function. The input is high- and low-frequency ground-penetrating radar images, and the output is the corresponding pseudo-high-frequency image.

[0110] In one embodiment, such as Figure 6 As shown, the basic structure of Transformer high-frequency feature fusion is illustrated. It mainly consists of three parts: a multi-band feature embedding module, a multi-head attention extraction module, and a dynamic weight prediction module.

[0111] In one embodiment, such as Figure 6 The diagram illustrates the effect of the simulated data after steps 3 and 4, as well as the results of using steps 3 and 4 independently. The specific task is to fuse low-frequency (400MHz and 900MHz) simulated data with high-frequency (1600MHz) data. Figure 6 (a) and Figure 6 (b) shows the fusion results of 400MHz, 900MHz, and 1600MHz obtained using the module in step 3, respectively. Figure 6 (c) and Figure 6 (d) represents the fusion results of 400MHz, 900MHz, and 1600MHz obtained using module 4, respectively. Figure 6 (e) and Figure 6 (f) represents the fusion results of 400MHz, 900MHz and 1600MHz obtained by using modules in steps 3 and 4, respectively. Figure 6 (a) and Figure 6 (b) shows the results of fusing high- and low-frequency data using the step 4 module alone. Figure 6 (c) andFigure 6 (d) shows the results using only the step 3 module and the high and low frequency data. Figure 6 (e) and Figure 7 Figure (f) shows the result of using steps 3 and 4 together. Using only step 4, the model fails to learn the mapping relationship well, resulting in the frequency transition band between shallow and deep layers being almost completely submerged in noise, while the middle soil layer remains blurred. Although the result improves as the frequency difference narrows, significant blocky noise remains in the background, and the overall quality of the fused image is not high. Using only step 3, while the hyperbolic contour of the underground target can be roughly reconstructed, the shallow image suffers from severe resolution deficiency, and soil background clutter interference is significant. The image fused using steps 3 and 4 shows a significant reduction in artifacts in the shallow region, and the visual effect is closer to the ideal high-frequency data.

[0112] In one embodiment, such as Figure 7 The results show the effect of the measured data after passing through steps 3 and 4, as well as the results after passing through steps 3 and 4 alone. The specific task is to fuse low-frequency (900MHz) and high-frequency (1600MHz) data. Figure 7 (a) shows the result of fusing 900MHz and 1600MHz using the module in step 3. Figure 7 (b) shows the result of fusing 900MHz and 1600MHz using the module in step 4. Figure 9 Image (c) shows the fusion result of 900MHz and 1600MHz obtained using modules 3 and 4. Using only the result from step 3 (a) enhances the detail representation of deep soil layers, but its reconstruction effect in shallow areas is not ideal. Using only the result from step 4 (b), while preserving the high-resolution fine roots of shallow layers, does not improve the resolution of deep soil layers. The fusion result combining steps 3 and 4 shows high visual consistency with real high-frequency data, especially in key soil stratification areas, where high-frequency texture recovery is significant, while also well preserving important fine root details in the shallow layer (0~5ns). This indicates that the multi-scene ground-penetrating radar multi-frequency fusion method not only improves resolution but also successfully leverages the penetration depth advantage of low-frequency signals to generate a comprehensive image containing richer geological structure information and reducing information loss.

[0113] Based on the same inventive concept, the embodiments of the present application also provide a ground penetrating radar data multi-frequency fusion device for implementing the above-mentioned ground penetrating radar data multi-frequency fusion method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more ground penetrating radar data multi-frequency fusion device embodiments provided below can refer to the limitations of the ground penetrating radar data multi-frequency fusion method described above, which will not be repeated here.

[0114] In one exemplary embodiment, as shown in Figure 10 A ground penetrating radar data multi-frequency fusion device is provided, comprising:

[0115] An acquisition module is configured to acquire multi-frequency ground penetrating radar data, wherein the multi-frequency ground penetrating radar data comprises ground penetrating radar data with at least two different frequency characteristics.

[0116] A preprocessing module is configured to preprocess the multi-frequency ground penetrating radar data to obtain preprocessed multi-frequency data.

[0117] A cross-frequency mapping module is configured to perform cross-frequency mapping on the preprocessed multi-frequency data using a Cycle-GAN network to obtain pseudo-high-frequency ground penetrating radar data.

[0118] An adaptive fusion module is configured to perform adaptive fusion on the pseudo-high-frequency ground penetrating radar data and actual high-frequency data using a Transformer fusion module to obtain a weighted fusion result.

[0119] In an exemplary embodiment, a computer device can be a server or a terminal, and its internal structure diagram can be as shown in Figure 10 The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store ground penetrating radar data multi-frequency fusion data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through network connection. The computer program is executed by the processor to implement a ground penetrating radar data multi-frequency fusion method.

[0120] Those skilled in the art can understand that,​ The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above method embodiments.

[0121] In an exemplary embodiment, a computer readable storage medium is provided, storing a computer program, the computer program being executed by a processor to implement the above method embodiments.

[0122] In an exemplary embodiment, a computer program product is provided, including a computer program, the computer program being executed by a processor to implement the above method embodiments.

[0123] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0124] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to a memory, a database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile and a volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0125] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0126] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0127] The principles and implementation modes of the present application are described by applying specific examples in the present application. The above-mentioned embodiments are only used to help understand the method and its core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A method for multi-frequency fusion of ground-penetrating radar data, characterized in that, The ground-penetrating radar data multi-frequency fusion method includes: Acquire multi-frequency ground-penetrating radar data; the multi-frequency ground-penetrating radar data includes ground-penetrating radar data with at least two different frequency characteristics; The multi-frequency ground-penetrating radar data is preprocessed to obtain preprocessed multi-frequency data. The preprocessed multi-frequency data is cross-frequency mapped using a Cycle-GAN network to obtain pseudo-high-frequency ground-penetrating radar data. The pseudo-high frequency ground-penetrating radar data and the actual high frequency data are adaptively fused using the Transformer fusion module to obtain a weighted fusion result. The calculation process of the loss function of the Transformer fusion module specifically includes: determining the prior weights of high-frequency components and spatial gradients based on pseudo-high-frequency ground-penetrating radar data and actual high-frequency data; and performing a short-time Fourier transform on the input signal to obtain its time-frequency representation. Define high-frequency prior weights Preset high frequency band Internal energy; This is the lower limit of the effective frequency band. The upper limit of the effective frequency band, Indicates the actual frequency of the data; A second-order difference operator is used to detect drastic changes in spatial gradient. This operator generates spatial gradient prior weights by convolving with the image. The joint weights are determined based on the prior weights of high-frequency components and the prior weights of spatial gradients. The weighted pixel-level error and structural similarity loss are determined based on the joint weights.

2. The multi-frequency fusion method for ground-penetrating radar data according to claim 1, characterized in that, The preprocessing includes zero-time correction, removal of direct waves, time-varying gain processing, and amplitude normalization.

3. The multi-frequency fusion method for ground-penetrating radar data according to claim 1, characterized in that, The loss function of the Cycle-GAN network is a weighted sum of bidirectional adversarial loss, cycle consistency loss, and multi-scale structural similarity loss.

4. The multi-frequency fusion method for ground-penetrating radar data according to claim 1, characterized in that, The loss function of the Transformer fusion module includes weighted pixel-level error and structural similarity loss.

5. The multi-frequency fusion method for ground-penetrating radar data according to claim 1, characterized in that, The Transformer fusion module includes a multi-band feature embedding module, a multi-head attention extraction module, and a dynamic weight prediction module connected in sequence.

6. A multi-frequency fusion device for ground-penetrating radar data, characterized in that, The ground-penetrating radar data multi-frequency fusion device includes: An acquisition module is used to acquire multi-frequency ground-penetrating radar data; the multi-frequency ground-penetrating radar data includes ground-penetrating radar data with at least two different frequency characteristics; The preprocessing module is used to preprocess multi-frequency ground-penetrating radar data to obtain preprocessed multi-frequency data. The cross-frequency mapping module is used to perform cross-frequency mapping on the preprocessed multi-frequency data using a Cycle-GAN network to obtain pseudo-high-frequency ground-penetrating radar data. An adaptive fusion module is used to adaptively fuse the pseudo-high frequency ground-penetrating radar data and the actual high frequency data using the Transformer fusion module to obtain a weighted fusion result. The calculation process of the loss function of the Transformer fusion module specifically includes: determining the prior weights of high-frequency components and spatial gradients based on pseudo-high-frequency ground-penetrating radar data and actual high-frequency data; and performing a short-time Fourier transform on the input signal to obtain its time-frequency representation. Define high-frequency prior weights Preset high frequency band Internal energy; This is the lower limit of the effective frequency band. The upper limit of the effective frequency band, Indicates the actual frequency of the data; A second-order difference operator is used to detect drastic changes in spatial gradient. This operator generates spatial gradient prior weights by convolving with the image. The joint weights are determined based on the prior weights of high-frequency components and the prior weights of spatial gradients. The weighted pixel-level error and structural similarity loss are determined based on the joint weights.

7. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the ground-penetrating radar data multi-frequency fusion method according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-frequency fusion method for ground-penetrating radar data as described in any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the multi-frequency fusion method for ground-penetrating radar data as described in any one of claims 1-5.

Images