Underground medium electrical characteristic inversion algorithm based on ground penetrating radar signal time-frequency characteristics

Abstract

The invention discloses an underground medium electrical characteristic inversion algorithm based on ground penetrating radar signal time-frequency characteristics. The algorithm is based on an encoder-decoder jump connection structure, and the dielectric constant and the conductivity are accurately inverted by fully combining the change of frequency domain characteristics in the attenuation process of radar waves. Experimental results show that the algorithm shows excellent inversion capability on the electrical characteristics of the underground medium under the non-uniform soil condition.

Description

Technical Field

[0001] The invention relates to the field of underground medium electrical characteristic inversion using two-dimensional ground penetrating radar, and more specifically to an underground medium electrical characteristic inversion algorithm based on the time-frequency characteristics of ground penetrating radar signals. Background Art

[0002] Ground penetrating radar is an efficient, non-destructive and high-precision geophysical detection method. If you want to analyze and interpret underground structures, you need to reconstruct the distribution and properties of the underground medium from the ground penetrating radar data, that is, to deduce the physical parameters and structural information of the underground medium from the measured ground penetrating radar echo signal. This process is called inversion.

[0003] With the advancement of deep learning, in order to further improve the efficiency and accuracy of inversion, ground penetrating radar inversion research based on deep learning has become popular. During ground penetrating radar detection, changes in the conductivity of the underground medium affect the attenuation of the electromagnetic wave signal, and changes in the dielectric constant affect the reflection, refraction and diffraction of the electromagnetic wave. Exploring the mapping relationship between radar signals and the dielectric properties of underground media has become the key to solving the ground penetrating radar inversion problem. However, there is currently a lack of inversion research on the conductivity of underground media, and the attenuation mechanism of radar signals has not been fully combined with the idea of ​​deep learning. Summary of the invention

[0004] In view of this, in order to further explore the application of frequency domain characteristics of ground penetrating radar signals in deep learning algorithms, the present invention proposes an underground medium electrical characteristics inversion algorithm based on the time-frequency characteristics of ground penetrating radar signals. First, the center frequency and variance of the characteristic parameters of electromagnetic wave attenuation are derived and calculated using the centroid frequency shift method. Secondly, the peak frequency shift method is used to derive and calculate the peak frequency and amplitude of the characteristic parameters of electromagnetic wave attenuation. Finally, the time-frequency features are fused, and a network composed of an encoder and a decoder connected by jumps is used to complete the inversion of the electrical characteristics of the underground medium. This invention can effectively improve the classification accuracy of underground targets of ground penetrating radar and provide a new idea for the inversion of ground penetrating radar data.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] An underground medium electrical property inversion algorithm based on the time-frequency characteristics of ground penetrating radar signals, characterized by comprising:

[0007] S1: Construct signal frequency domain features. The specific steps are as follows:

[0008] S11: Finding characteristic parameters of electromagnetic wave attenuation: The quality factor Q is the key parameter to mathematically quantify the attenuation effect. The result of estimating Q using the centroid frequency shift method is: where f S and are the center frequency and variance of the source wavelet, respectively, and f R is the center frequency of the received amplitude spectrum. From the expression of Q, it can be observed that the center frequency and variance can reflect the change of Q value. The result of estimating Q using the peak frequency shift method is where f p is the peak frequency, f m is the main frequency. From the expression of Q, it can be observed that the peak frequency f p It can reflect the change of Q value and is related to a specific amplitude B(f p ). Therefore, the change in amplitude can also represent the change in Q value.

[0009] S12: Calculate the frequency domain characteristic parameters derived in S11: Use bilinear interpolation to convert the two-dimensional radar data into a size of (64, 512), and number each column of data (taking the first column as an example) as y 0 -y 511 And do the following processing: When calculating the frequency domain information of the data for the i-th time (i=0), the data y 0 -y 31 Wavelet transform, extract the center frequency, variance, amplitude and peak frequency; when calculating the frequency domain information of the data for the i-th time (i ≥ 1), slide down with a time window of 16 and a step size of 1, and perform the following operations on the data y in turn. i+15 -y i+30 Wavelet transform is performed, and the center frequency, variance, amplitude and peak frequency are extracted. S13: Frequency domain feature fusion: The four frequency domain information calculated by S12 are combined into a four-channel frequency domain feature data. The arrangement of these four channels can be expressed as [F c ,F v ,F p ,F a ], where F c is the center frequency of the radar data, F v is the variance of radar data, F p is the peak frequency of radar data, F a is the amplitude of the radar data.

[0010] S2: Time-frequency feature fusion: The time domain signal of the radar data is processed by the encoder. After four convolutions and one downsampling, the convolution result has the same size as each channel of the frequency domain feature information. After the two are fused, they continue to participate in the subsequent processing of the encoder.

[0011] S3: Output the inversion results of underground medium electrical properties: The encoder in S2 is connected to a decoder through jump connection to form an inversion network to achieve accurate inversion of dielectric constant and conductivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0013] Figure 1 The accompanying drawing is a flow chart of an underground medium electrical property inversion algorithm based on the time-frequency characteristics of ground penetrating radar signals provided by the present invention.

[0014] Figure 2 The attached figure is a schematic diagram of frequency domain feature data processing for calculating two-dimensional ground penetrating radar signals.

[0015] Figure 3 The attached figure is a schematic diagram of the time-frequency feature fusion module.

[0016] Figure 4 The attached figure is a model structure diagram of the inversion algorithm for underground medium electrical properties using ground penetrating radar.

[0017] Specific implementation method The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] See attached Figure 1 The present invention discloses an inversion algorithm for underground medium electrical characteristics based on the time-frequency characteristics of ground penetrating radar signals. The algorithm includes: S1: constructing signal frequency domain characteristics. The specific steps are as follows:

[0019] S11: Finding characteristic parameters of electromagnetic wave attenuation: The quality factor Q is the key parameter to mathematically quantify the attenuation effect. The result of estimating Q using the centroid frequency shift method is: where f S and are the center frequency and variance of the source wavelet, respectively, and f R is the center frequency of the received amplitude spectrum. From the expression of Q, it can be observed that the center frequency and variance can reflect the change of Q value. The result of estimating Q using the peak frequency shift method is where f p is the peak frequency, f m is the main frequency. From the expression of Q, it can be observed that the peak frequency f p It can reflect the change of Q value and is related to a specific amplitude B(f p ). Therefore, the change in amplitude can also represent the change in Q value.

[0020] S12: Calculate the frequency domain characteristic parameters derived in S11: Use bilinear interpolation to convert the two-dimensional radar data into a size of (64, 512), and number each column of data (taking the first column as an example) as y 0 -y 511 And do the following processing: When calculating the frequency domain information of the data for the i-th time (i=0), the data y 0 -y 31 Wavelet transform, extract the center frequency, variance, amplitude and peak frequency; when calculating the frequency domain information of the data for the i-th time (i ≥ 1), slide down with a time window of 16 and a step size of 1, and perform the following operations on the data y in turn. i+15 -y i+30 Wavelet transform and extract center frequency, variance, amplitude and peak frequency. Figure 2 The figure shows the process of calculating the frequency domain characteristic data of two-dimensional ground penetrating radar signal.

[0021] S13: Frequency domain feature fusion: The four frequency domain information calculated by S12 are combined into a four-channel frequency domain feature data. The arrangement of these four channels can be expressed as [F c ,F v ,F p ,F a ], where F c is the center frequency of the radar data, F v is the variance of radar data, F p is the peak frequency of radar data, F a is the amplitude of the radar data.

[0022] S2: Time-frequency feature fusion: The time domain signal of the radar data is processed by the encoder. After four convolutions and one downsampling, the convolution result has the same size as each channel of the frequency domain feature information. After the two are fused, they continue to participate in the subsequent processing of the encoder. Figure 3 Shown is a schematic diagram of the time-frequency feature fusion module.

[0023] S3: Output the inversion results of underground medium electrical properties: The encoder in S2 is connected to a decoder through jump connection to form an inversion network to achieve accurate inversion of dielectric constant and conductivity. Figure 4 Shown is the model structure diagram of the ground penetrating radar underground medium electrical properties inversion algorithm.

[0024] The following specifically describes the origin of the encoder-decoder network and skip connection provided by the present invention.

[0025] In early sequence processing tasks, recurrent neural networks or convolutional neural networks are usually used to process sequence data. However, these networks have the problem of gradient vanishing or gradient exploding when processing long sequences, which makes it difficult for the model to learn long-distance dependencies. In order to solve the problem of long sequence processing, researchers proposed an encoder-decoder architecture. In this structure, the encoder is responsible for encoding the input sequence into a fixed-length vector (called a context vector or encoded representation), which contains all the information of the input sequence. The decoder generates an output sequence based on this context vector. With the development of deep learning technology, encoder-decoder networks have been widely used in other fields, such as image processing, audio processing, etc. For example, in the image description task, the encoder is responsible for extracting the features of the image, and the decoder generates a natural language sentence describing the image based on these features.

[0026] The skip connection is a structure originally proposed in the residual network. It increases the flow of information and reduces the gradient vanishing problem by establishing direct connections between adjacent layers of the network, thereby allowing the network to be trained more deeply. The skip connection usually includes an identity mapping and an optional linear projection. The skip connection facilitates the flow of information between different layers, allowing the model to better learn and propagate gradients. In deep networks, the skip connection can reduce the gradient vanishing problem and allow the network to be trained more deeply.

[0027] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An inversion algorithm for underground dielectric electrical properties based on the time-frequency characteristics of ground penetrating radar signals, characterized in that: include: S1: Construct signal frequency domain features, specific steps as follows: S11: Finding characteristic parameters of electromagnetic wave attenuation: The quality factor Q is the key parameter to mathematically quantify the attenuation effect. The result of estimating Q using the centroid frequency shift method is: where f S and are the center frequency and variance of the source wavelet, respectively, and f R is the center frequency of the received amplitude spectrum. From the expression of Q, it can be observed that the center frequency and variance can reflect the change of Q value. The result of estimating Q using the peak frequency shift method is where f p is the peak frequency, f m is the main frequency. From the expression of Q, it can be observed that the peak frequency f p It can reflect the change of Q value and is related to a specific amplitude B(f p ). Therefore, the change in amplitude can also represent the change in Q value. S12: Calculate the frequency domain characteristic parameters derived in S11: Use bilinear interpolation to convert the two-dimensional radar data into a size of (64, 512), and number each column of data (taking the first column as an example) as y0-y 511 And do the following processing: When calculating the frequency domain information of the data for the i-th time (i=0), for the data y0-y 31 Wavelet transform, extract the center frequency, variance, amplitude and peak frequency; when calculating the frequency domain information of the data for the i-th time (i ≥ 1), slide down with a time window of 16 and a step size of 1, and perform the following operations on the data y in turn. i+15 -y i+30 Wavelet transform and extract center frequency, variance, amplitude and peak frequency. S13: Frequency domain feature fusion: The four frequency domain information calculated by S12 are combined into a four-channel frequency domain feature data. The arrangement of these four channels can be expressed as [F c ,F v ,F p ,F a ], where F c is the center frequency of the radar data, F v is the variance of radar data, F p is the peak frequency of radar data, F a is the amplitude of the radar data. S2: Time-frequency feature fusion: The time domain signal of the radar data is processed by the encoder. After four convolutions and one downsampling, the convolution result has the same size as each channel of the frequency domain feature information. After the two are fused, they continue to participate in the subsequent processing of the encoder. S3: Output the inversion results of underground medium electrical properties: The encoder in S2 is connected to a decoder through jump connection to form an inversion network to achieve accurate inversion of dielectric constant and conductivity.

Images