A method for predicting the dielectric constant of the Martian shallow surface layer based on an improved convolutional neural network

Abstract

The present application provides a method for predicting the dielectric constant of the shallow layer of Mars using an improved convolutional neural network, which relates to the fields of geological exploration and planetary exploration. The method includes: obtaining and preprocessing Mars subsurface detection radar data; extracting Mars subsurface detection radar A-scan data; constructing a one-dimensional convolutional neural network based on an attention mechanism and cascade; training the one-dimensional convolutional neural network using the constructed Mars subsurface geological model and Mars subsurface detection radar A-scan data; and predicting the dielectric constant value of the shallow subsurface layer of Mars using the trained one-dimensional convolutional neural network in combination with the preprocessed Mars subsurface detection radar data. The method proposed in the present application uses a data-driven improved one-dimensional deep convolutional neural network to replace the traditional manual judgment of hyperbola fitting in ground-penetrating radar signals, thereby realizing a method for autonomously predicting the dielectric properties of the shallow subsurface layer of Mars.

Description

Technical Field

[0001] The present application relates to the fields of geological exploration and planetary exploration, and in particular to a method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network. Background Art

[0002] Ground-penetrating radar (GPR) technology has been widely used to detect subsurface structures and identify targets, including unexploded ordnance detection, roadbed assessment, and other civil engineering and environmental applications. GPR transmits electromagnetic waves downward, receives reflected signals from subsurface objects, and combines a series of electromagnetic wave trajectories to form high-resolution image data. Subsurface objects appear as hyperbolic curves in GPR's B-scan (two-dimensional profile) data. There are various methods for estimating the dielectric properties of subsurface media, including the hyperbolic method, full waveform inversion, and the center point method. Hyperbolic fitting is the most commonly used calculation method. Using the hyperbolic method to estimate the dielectric constant of the background medium generally assumes that the target object exists within a relatively homogeneous background medium. However, since objects are often placed within a non-homogeneous background medium in reality, the reflected signals from the target object exhibit non-standard hyperbolic characteristics in the B-scan, resulting in significant errors in the estimated results.

[0003] During exoplanet exploration of exoplanets such as Mars and the Moon, radar instruments typically carry air-coupled antennas. The signals reflected from shallow subsurface objects on Mars exhibit a non-standard hyperbolic pattern, and these radar images cannot directly reflect the characteristics of the target object and the background medium. Therefore, data interpretation and inversion are crucial. Accurate and precise data inversion and interpretation rely heavily on researchers' experience and field verification. However, the sheer volume of subsurface radar data acquired during Mars exploration makes manual interpretation prohibitively expensive, and field verification is also challenging.

[0004] Therefore, there is an urgent need for an autonomous Mars subsurface radar data inversion and interpretation method, whose main application is the automatic evaluation of the dielectric constant of the shallow surface of Mars. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for predicting the dielectric constant of the shallow surface of Mars using an improved convolutional neural network in order to solve the problem that manual interpretation of Mars subsurface detection radar data is too costly and has low feasibility.

[0006] The above-mentioned purpose of this application is achieved through the following technical solutions:

[0007] S1: Acquire Mars subsurface radar data and perform preprocessing;

[0008] S2: Extracting Mars subsurface radar A-scan data;

[0009] S3: Construct a one-dimensional convolutional neural network based on attention mechanism and cascade;

[0010] S4: Train a one-dimensional convolutional neural network using a geological model of the dielectric constant of the Martian regolith and A-scan data from the Martian subsurface radar.

[0011] S5: The dielectric constant of the shallow subsurface layer of Mars is predicted by combining the trained one-dimensional convolutional neural network with the preprocessed Mars subsurface radar data.

[0012] Optionally, the preprocessing includes: data reading and correction, DC removal, background noise removal, gain compensation, bandpass filtering, and partial correction signal removal.

[0013] Optionally, step S1 includes:

[0014] S11: Acquire Mars subsurface radar data;

[0015] Data reading and correction: Delete the self-test signal in the Mars subsurface detection radar data;

[0016] S12: DC removal: removes DC low-frequency components from Mars subsurface radar data;

[0017] S13: Background noise removal: Subtract the average value of all Mars subsurface radar data in each segment from the Mars subsurface radar data in the segment;

[0018] S14: Gain compensation: Perform geometric diffusion compensation on the Mars subsurface radar data, and then perform nonlinear gain;

[0019] S15: Bandpass filtering: Filtering of Mars subsurface radar data in a specific frequency band;

[0020] S16: Partial correction signal removal: Use the Tukey window function to filter out the residual direct wave signals of the Mars subsurface detection radar data.

[0021] Optionally, step S2 includes:

[0022] S21: Establish a geological model of the dielectric constant of the Martian regolith;

[0023] S22: Based on the dielectric constant geological model of the Martian regolith, the open source software gprMax for ground-penetrating radar forward modeling is used to perform random inhomogeneous medium forward modeling to obtain numerically simulated Martian subsurface ground-penetrating radar A-scan data.

[0024] Optionally, step S4 includes:

[0025] S41: Use numerically simulated Martian subsurface ground-penetrating radar A-scan data and the corresponding Martian regolith dielectric constant geological model as deep learning training data and labels, respectively, to construct a dataset;

[0026] S42: Train and validate the one-dimensional convolutional neural network using the dataset.

[0027] Optionally, the input data size of the one-dimensional convolutional neural network is set to 800×500, where 800 represents the number of sampling points per channel and 500 represents the number of A-scan channels;

[0028] The one-dimensional convolutional neural network includes: a convolution module, a self-attention layer, a cascade layer Cascade, a fifth convolution layer, an upsampling layer, a deconvolution layer Deconv1, a deconvolution layer Deconv2, a deconvolution layer Deconv3, a deconvolution layer Deconv4, a deconvolution layer Deconv5, and an output layer;

[0029] The convolution module, the self-attention layer, the cascade layer, the fifth convolution layer, and the upsampling layer are connected in sequence;

[0030] The convolution module, the deconvolution layer Deconv1, and the deconvolution layer Deconv2 are connected in sequence;

[0031] The upsampling layer and the deconvolution layer Deconv2 are both connected to the deconvolution layer Deconv3;

[0032] The deconvolution layer Deconv3, the deconvolution layer Deconv4, the deconvolution layer Deconv5, and the output layer are connected in sequence.

[0033] An electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs a method for predicting the dielectric constant of the shallow layer of Mars using an improved convolutional neural network.

[0034] A computer-readable storage medium stores instructions. When the instructions are executed, a method for predicting the dielectric constant of the shallow layer of Mars using an improved convolutional neural network is executed.

[0035] The beneficial effects of the technical solution provided by this application are:

[0036] A one-dimensional deep convolutional neural network based on attention mechanism and cascade is constructed. As a result, deep learning methods can be used to replace the traditional hyperbola fitting method to invert the dielectric constant, avoiding the time-consuming and labor-intensive phenomenon of over-reliance on the experience of interpreters, improving work efficiency, and realizing automated interpretation. It is also suitable for the rapid and automatic inversion and interpretation of dielectric properties in future planetary exploration. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present application will be further described below with reference to the accompanying drawings and embodiments, in which:

[0038] Figure 1 is a flow chart in an embodiment of the present application;

[0039] Figure 2 is a prediction evaluation index diagram in the embodiment of the present application;

[0040] Figure 3 is a schematic diagram of the structure of an electronic device in an embodiment of the present application;

[0041] Figure 4 This is a one-dimensional convolutional neural network structure diagram in an embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to have a clearer understanding of the technical features, purposes and effects of this application, the specific implementation methods of this application are now described in detail with reference to the accompanying drawings.

[0043] An embodiment of the present application provides a method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network.

[0044] Please refer to Figure 1 , Figure 1 : is a flowchart of a method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network in an embodiment of the present application, including:

[0045] S1: Acquire Mars subsurface radar data and perform preprocessing;

[0046] The preprocessing includes: data reading and correction, DC removal, background noise removal, gain compensation, bandpass filtering, and partial correction signal removal.

[0047] Step S1 includes:

[0048] S11: Acquire Mars subsurface radar data;

[0049] Data reading and correction: Delete the self-test signal in the Mars subsurface detection radar data;

[0050] Specifically, each segment of high-frequency channel data contains self-check signals, which prevents us from directly extracting valid information from the image. Therefore, removing these self-check signals is necessary. Furthermore, any remaining direct wave signals are ultimately removed through partial correction signal removal.

[0051] S12: DC removal: removes DC low-frequency components from Mars subsurface radar data;

[0052] Specifically, the received radar signal contains a low-frequency component with a frequency close to DC. If it is not removed, it will have an adverse effect on the subsequent gain compensation processing results, such as signal distortion and signal average shift.

[0053] S13: Background noise removal: Subtract the average value of all Mars subsurface radar data in each segment from the Mars subsurface radar data in the segment;

[0054] Specifically, due to interference from the instrument itself and the Martian environment, the Mars subsurface radar signal contains a lot of background noise, which can adversely affect the data (affecting interpretation and interpretation). This step removes unwanted background noise by subtracting the average of all data within each Mars subsurface radar data segment.

[0055] S14: Gain compensation: Perform geometric diffusion compensation on the Mars subsurface radar data, and then perform nonlinear gain;

[0056] Specifically, the electromagnetic wave signals received by the receiving antenna of the Mars subsurface ground-penetrating radar will be affected by geometric diffusion and dielectric loss, resulting in weaker reflected signal energy from deeper targets. Therefore, it is necessary to perform geometric diffusion compensation on the Mars subsurface detection radar data and then perform nonlinear gain.

[0057] S15: Bandpass filtering: Filtering of Mars subsurface radar data in a specific frequency band;

[0058] Specifically, the bandwidth of the high-frequency channel of the Mars subsurface radar is limited to the range of 0.45–2.15 GHz. This step sets the frequency range of the bandpass filter to 0.45–2.15 GHz, aiming to retain signals within this target frequency band and filter out signals in other frequency bands.

[0059] S16: Partial correction signal removal: Use the Tukey window function to filter out the residual direct wave signals of the Mars subsurface detection radar data.

[0060] S2: Extracting Mars subsurface radar A-scan data;

[0061] Step S2 includes:

[0062] S21: Establish a geological model of the dielectric constant of the Martian regolith;

[0063] S22: Based on the dielectric constant geological model of the Martian regolith, the open source software gprMax for ground-penetrating radar forward modeling is used to perform random inhomogeneous medium forward modeling to obtain numerically simulated Martian subsurface ground-penetrating radar A-scan data.

[0064] Specifically, the actual Martian geology and topography are extremely complex. To date, no one has actually set foot on Mars, leading to a variety of differing viewpoints on the study and interpretation of the Martian subsurface. Using deep learning networks to predict dielectric constants requires a large amount of training data. To address this issue, a large amount of simulation data is proposed for network training. Previous research suggests that the Martian regolith may exhibit subsurface stratification. Based on this, a series of geological models of the Martian regolith were simulated, primarily to establish complex geological models with distinct upper and lower stratification and undulating interfaces. These models include models with fine sand-like particles and no undulating interfaces, and models with fine sand particles but varying degrees of undulation. Based on these geological models, the open-source GPR forward modeling software gprMax was used to perform random inhomogeneous media forward modeling, acquiring a series of GPR A-scan data.

[0065] S3: Construct a one-dimensional convolutional neural network based on attention mechanism and cascade;

[0066] S4: Train a one-dimensional convolutional neural network using a geological model of the dielectric constant of the Martian regolith and A-scan data from the Martian subsurface radar.

[0067] Step S4 includes:

[0068] S41: Use numerically simulated Martian subsurface ground-penetrating radar A-scan data and the corresponding Martian regolith dielectric constant geological model as deep learning training data and labels, respectively, to construct a dataset;

[0069] S42: Train and validate the one-dimensional convolutional neural network using the dataset.

[0070] The input data size of the one-dimensional convolutional neural network is set to 800×500, where 800 represents the number of sampling points per channel and 500 represents the number of A-scan channels;

[0071] The overall architecture of a one-dimensional convolutional neural network is as follows Figure 4 As shown in the figure, the input A-scan data first passes through the encoding layer (a series of convolutional layers, attention mechanism layers, and cascade layers), and finally passes through the decoding layer (a series of deconvolution layers) to obtain the corresponding dielectric constant prediction value.

[0072] The one-dimensional convolutional neural network includes: a convolution module, a self-attention layer, a cascade layer Cascade, a fifth convolution layer, an upsampling layer, a deconvolution layer Deconv1, a deconvolution layer Deconv2, a deconvolution layer Deconv3, a deconvolution layer Deconv4, a deconvolution layer Deconv5, and an output layer;

[0073] The convolution module, the self-attention layer, the cascade layer, the fifth convolution layer, and the upsampling layer are connected in sequence;

[0074] The convolution module, the deconvolution layer Deconv1, and the deconvolution layer Deconv2 are connected in sequence;

[0075] The upsampling layer and the deconvolution layer Deconv2 are both connected to the deconvolution layer Deconv3;

[0076] The deconvolution layer Deconv3, the deconvolution layer Deconv4, the deconvolution layer Deconv5, and the output layer are connected in sequence.

[0077] Specifically, the Mars subsurface ground penetrating radar moves 0.05m each time along the horizontal side line of the surface to obtain the Mars subsurface ground penetrating radar A-scan data, and a total of 500 traces are obtained. The data to be trained is sent to a one-dimensional convolutional neural network based on the attention mechanism (Self-Attention) and cascade (Cascade). This network directly learns the nonlinear mapping relationship between the Mars subsurface ground penetrating radar A-scan data and the dielectric constant. The entire data set is divided into training set and test set at a ratio of 99:1. The training set is used for network training, and the test set is used for network verification. During the training phase of the network, the weights of the network are iteratively updated until convergence. In this method, the network iterates a total of 24 epochs (epoch refers to the number of times the model is trained on the entire training set), and the learning rate is 0.0001. After the network converges, the network enters the verification phase, and the above-mentioned verification set is used to evaluate the accuracy of the network. As Figure 2 As shown, the prediction evaluation metric is measured using R², a commonly used metric in deep learning to measure the fit of a model to real-world data. R² values ​​range from 0 to 1, with larger values ​​indicating better fit. The blue line represents the actual labels (i.e., dielectric constants) of the A-scans in the validation set, while the red line represents the dielectric constants predicted by the network for the A-scans in the validation set, achieving a 99% fit.

[0078] S5: The dielectric constant of the shallow subsurface layer of Mars is predicted by combining the trained one-dimensional convolutional neural network with the preprocessed Mars subsurface radar data.

[0079] This application also discloses an electronic device. Figure 3 , Figure 3 Schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.

[0080] The communication bus 502 is used to implement the connection and communication between these components.

[0081] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.

[0082] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0083] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for loading by a processor to execute the above-mentioned improved convolutional neural network method for predicting the dielectric constant of the shallow surface layer of Mars.

[0084] The above are merely exemplary embodiments of the present disclosure and are not intended to limit the scope of the present disclosure. In other words, any equivalent variations and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Those skilled in the art will readily conceive of other embodiments of the present disclosure after considering the disclosure and the practical implications thereof.

[0085] This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not described herein. The description and examples are to be considered as exemplary only, and the scope and spirit of the present disclosure are to be defined by the claims.

Claims

1. A method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network, characterized in that: The method comprises the following steps: S1: Acquire Mars subsurface radar data and perform preprocessing; S2: Extracting Mars subsurface radar A-scan data; S3: Construct a one-dimensional convolutional neural network based on attention mechanism and cascade; The input data size of the one-dimensional convolutional neural network is set to 800×500, where 800 represents the number of sampling points per channel and 500 represents the number of A-scan channels; The one-dimensional convolutional neural network includes: a convolution module, a self-attention layer, a cascade layer Cascade, a fifth convolution layer, an upsampling layer, a deconvolution layer Deconv1, a deconvolution layer Deconv2, a deconvolution layer Deconv3, a deconvolution layer Deconv4, a deconvolution layer Deconv5, and an output layer; The convolution module, the self-attention layer, the cascade layer, the fifth convolution layer, and the upsampling layer are connected in sequence; The convolution module, the deconvolution layer Deconv1, and the deconvolution layer Deconv2 are connected in sequence; The upsampling layer and the deconvolution layer Deconv2 are both connected to the deconvolution layer Deconv3; The deconvolution layer Deconv3, the deconvolution layer Deconv4, the deconvolution layer Deconv5, and the output layer are connected in sequence; S4: Train a one-dimensional convolutional neural network using a geological model of the dielectric constant of the Martian regolith and A-scan data from the Martian subsurface radar. S5: The dielectric constant of the shallow subsurface layer of Mars is predicted by combining the trained one-dimensional convolutional neural network with the preprocessed Mars subsurface radar data.

2. The method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network according to claim 1, wherein: The preprocessing includes: data reading and correction, DC removal, background noise removal, gain compensation, bandpass filtering, and partial correction signal removal.

3. The method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network according to claim 2, wherein: Step S1 includes: S11: Acquire Mars subsurface radar data; Data reading and correction: Delete the self-test signal in the Mars subsurface detection radar data; S12: DC removal: removes DC low-frequency components from Mars subsurface radar data; S13: Background noise removal: Subtract the average value of all Mars subsurface radar data in each segment from the Mars subsurface radar data in the segment; S14: Gain compensation: Perform geometric diffusion compensation on the Mars subsurface radar data, and then perform nonlinear gain; S15: Bandpass filtering: Filtering of Mars subsurface radar data in a specific frequency band; S16: Partial correction signal removal: Use the Tukey window function to filter out the residual direct wave signals of the Mars subsurface detection radar data.

4. The method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network according to claim 1, wherein: Step S2 includes: S21: Establish a geological model of the dielectric constant of the Martian regolith; S22: Based on the dielectric constant geological model of the Martian regolith, the open source software gprMax for ground-penetrating radar forward modeling is used to perform random inhomogeneous medium forward modeling to obtain numerically simulated Martian subsurface ground-penetrating radar A-scan data.

5. The method for predicting the dielectric constant of the shallow surface layer of Mars using an improved convolutional neural network according to claim 4, wherein: Step S4 includes: S41: Use numerically simulated Martian subsurface ground-penetrating radar A-scan data and the corresponding Martian regolith dielectric constant geological model as deep learning training data and labels, respectively, to construct a dataset; S42: Train and validate the one-dimensional convolutional neural network using the dataset.

6. An electronic device, characterized in that: The electronic device comprises a processor (501), a memory (505), a user interface (503) and a network interface (504), wherein the memory (505) is used to store instructions, the user interface (503) and the network interface (504) are used to communicate with other devices, and the processor (501) is used to execute the instructions stored in the memory (505) so that the electronic device executes the method according to any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 5 is executed.

Images