A dual-scale data fusion imaging method for Mars radar car data

Abstract

The present application belongs to the technical field of Mars radar vehicle data processing, and particularly relates to a double-scale data fusion imaging method for Mars radar vehicle data, which fuses RIMFAX and RoPeR double-source data to realize collaborative imaging of shallow precision and deep thickness in a depth range of more than 100 meters. Through grid alignment and Gaussian weight transition, stitching artifacts are eliminated, and the processing efficiency and imaging quality are significantly improved. The present application can overcome the defects in the existing Mars radar data processing, such as the single load being unable to consider the detection depth and resolution, and the double-source heterogeneous data being difficult to realize collaborative imaging, and provides a double-scale data fusion imaging method. Through differential filtering, unified grid registration and depth domain weighted fusion of RIMFAX shallow high-resolution data and RoPeR deep large-penetration data, integrated imaging of Mars shallow structure from shallow fine to deep profile is realized.

Description

Technical Field

[0001] This invention belongs to the field of Mars radar rover data processing technology, specifically relating to a dual-scale data fusion imaging method for Mars radar rover data. Background Technology

[0002] Mars rovers are core exploration platforms for revealing Martian subsurface structures and searching for water ice resources. However, their onboard radar payloads struggle to simultaneously meet the dual requirements of high resolution on the shallow surface and deep penetration during exploration. The RIMFAX radar, operating at a higher frequency, provides detailed depictions of shallow structures but has limited depth. RIMFAX stands for Radar Imager for Mars' Subsurface Exploration, a ground-penetrating radar payload carried by NASA's Perseverance rover. The RoPeR radar, with its low-frequency channel, offers deep penetration but lacks sufficient resolution. Furthermore, due to the complexity of the Martian exploration environment, the strong attenuation and noise characteristics of radar signals, and the heterogeneity of data formats and inconsistent sampling grids, single radar data cannot fully reveal the continuous subsurface structure from the shallow surface to the deep layers. Therefore, research on dual-scale fusion imaging methods using dual-source radar data from Mars is essential for achieving integrated characterization of detailed shallow structures and deep stratigraphic information, and for providing multi-scale collaborative imaging support for the interpretation of shallow Martian geology.

[0003] Patent application CN104793203B proposes a joint processing method for multi-frequency and multi-channel data from ground-penetrating radar. This technology performs unified correction processing on data collected from different channels, ensuring spatial consistency among the channel data. Based on this, it performs frequency domain feature analysis on the signals, comprehensively expressing the information from different channels to form a fused radar profile result. This method can, to some extent, utilize the advantages of different frequency signals to improve the stability of the detection results. The drawbacks of this method are: its processing is mainly limited to data within the same radar system, lacking adaptability to the differences in characteristics between different types of radar; furthermore, this method focuses on data integration at a unified scale, without layered processing for different detection depths and resolutions, and still has shortcomings in multi-scale structural representation.

[0004] Patent application CN117687042A proposes a radar data fusion method for multi-sensor collaboration. This method uses spatial data acquired by multiple radars as a foundation, and through coordinate system transformation and position alignment, maps data from different sources to the same reference coordinate system, thereby integrating multi-radar information and expanding the overall detection range. The drawbacks of this method are: its core is data alignment and fusion at the spatial location level, primarily serving applications such as environmental perception, lacking in-depth processing of the physical response characteristics of the radar signals themselves; furthermore, this method does not distinguish the differences in detection capabilities between different radars and lacks a hierarchical processing mechanism for information at different depths, thus limiting its application in underground structure imaging.

[0005] Therefore, this invention provides a dual-scale data fusion imaging method for Mars radar rover data. Summary of the Invention

[0006] The purpose of this invention is to provide a dual-scale data fusion imaging method for Mars radar rover data, which can overcome the shortcomings of existing Mars radar data processing, such as the inability of a single payload to simultaneously consider detection depth and resolution, and the difficulty of coordinating imaging of dual-source heterogeneous data. This dual-scale data fusion imaging method achieves integrated imaging of the Martian shallow subsurface structure from the fine surface to the deep contour by implementing differentiated filtering, unified grid registration, and depth domain weighted fusion of RIMFAX shallow high-resolution data and RoPeR deep deep penetrating data.

[0007] The specific technical solution adopted by this invention is as follows: A dual-scale data fusion imaging method for Mars radar rover data includes the following steps: Step 1: Acquire data from the Mars Rover RIMFAX shallow radar and RoPeR deep radar; RoPeR stands for Mars Rover Penetrating Radar, a dual-band ground-penetrating radar carried by China's Zhurong Mars rover. Step 2: Read RIMFAX data using the segyio library in Python, set ignore_geometry=True to bypass the incomplete gather definition, and set the original gather matrix as follows. ,in The number of time sampling points, The number of channels; the original sampling interval is Lane spacing is Uniform downsampling is performed in both the time and spatial domains; lateral sampling step size. Take 8-15, and extract the step size vertically. Take values ​​from 2 to 5, and then downsample the matrix. Defined as: ; in , The equivalent time sampling interval after downsampling becomes: The equivalent track spacing becomes The highest frequency of the original signal should meet the following requirements. Extract the sampling interval dt; downsample the original gathers and store them as a two-dimensional data matrix. A shallow ; Step 3: Read RoPeR .2c data file by file according to the preset sampling point count of 1024, and reconstruct the gather matrix; the horizontal trace extraction step size is 1-3, and the vertical sampling extraction step size is 1-3. After batch reading, the data is horizontally stitched together to generate a deep data matrix. A deep The process is as follows: in The total number of samples corresponding to the file size in bytes; preset number of sampling points per track. Then the first The number of channels in each file is Reconstruct into a matrix; then adjust by horizontal step size. Take 1-3 and longitudinal step size Take 1-3 downsampling samples: ; All K The files are concatenated horizontally to obtain a deep data matrix: ; Step 4: Perform shallow data filtering and enhancement processing; Step 4 specifically includes the following steps: Step 401: Perform DC removal operation: Let matrix The elements are Total mean for: ; After removing DC: ; Step 402: Perform 2D Median Filtering: For window size (Odd number), define the neighborhood set: ; The filtered output is the median of all values ​​in the neighborhood: ; Shallow data retrieval , .

[0008] Step 403: Perform time-direction mean smoothing: A one-dimensional moving average is performed along the time axis, with a window length of [missing information]. (Odd number); Let the current sampling point be... Neighborhood range ,but: ; Mean smoothing is equivalent to convolution with a rectangular window; its frequency response is a sinc function, exhibiting low-pass filtering characteristics, and its cutoff frequency is... ; shallow layer ; Step 404: Perform automatic gain control based on Hilbert envelope: First, calculate the signal envelope. For the first... channel signal Its analytical signal is: ; Hilbert transform Defined as: ; The Hilbert transform in the frequency domain corresponds to a phase shift. Positive frequency and Negative frequency; envelope is the modulus of the analytic signal: ; AGC in the sliding window Internal calculation of average envelope: ; The signal after gain compensation is: ; in It is a small positive number; this operation normalizes the signal energy within a local time window, and the shallow AGC window length is taken as... .

[0009] Step 405: Perform strip removal operation: Let the first The horizontal mean of the time sampling points is: ; To remove the bands, simply subtract the mean: ; Step 406: Perform Gaussian smoothing: The two-dimensional Gaussian kernel is defined as: ; in The longitudinal standard deviation controls the smoothness of the time direction. The lateral standard deviation controls the smoothness between channels. The smoothing operation is convolution: ; shallow layer , Final output .

[0010] Step 5: Perform deep data filtering and enhancement processing; Step 5 specifically includes the following steps: Step 501: Perform DC removal operation, the formula is the same as in step 401; Step 501: Perform two-dimensional median filtering: increase the window size to... , The mathematical definition is consistent with step 402. Step 503: Perform time-direction mean smoothing: window length The formula is the same as in step 403.

[0011] Step 504: Perform Hilbert envelope AGC operation: Increase AGC window length to The envelope and gain formulas are the same as in step 404; Step 505: Perform strip removal operation: Same as step 405; Step 506: Perform Gaussian smoothing: Compared to step 406, the smoothing intensity is increased. , The formula is the same as in step 406; Step 507: Perform nonlinear amplitude compression: Use logarithmic compression to compress the amplitude range without changing the sign. .

[0012] Step 6: Perform time-depth transformation and coordinate generation; Specifically, in step 6: the time sampling interval is set to 0.8-1.2 ns, the shallow wave velocity to 0.08-0.12 m / ns, the deep wave velocity to 0.06-0.10 m / ns, and the channel spacing to 0.4-0.6 m; and the values ​​are calculated according to the formulas: Depth axis: Distance axis: ; Generate shallow depth axes respectively z shallow With depth axis z deep Distance axes are generated according to track number and track spacing respectively. x shallow and xdeep .

[0013] Step 7: Perform unified grid control registration; Specifically, in step 7: Define a uniform depth grid number of 800-1500 and a uniform distance grid number of 500-1000. The depth range is the smaller of the maximum depths of the two sets of data, and the distance range is the smaller of the maximum distances of the two sets of data. Two-dimensional linear interpolation is used. ; First, interpolate the data along the depth direction to a unified depth axis, then interpolate it along the distance direction to a unified distance axis to obtain a spatially aligned matrix. B shallow and B deep .

[0014] Step 8: Define the Gaussian weighting function for the depth domain; Specifically, in step 8: Transition depth z c Constructing with depth coordinate z as independent variable Shallow weight function: ; Deep weight function: ; Step 9: Perform point-by-point weighted summation of the fusion matrix F using the following formula; The fusion matrix F in step 9 is calculated by weighted summation point by point according to the following formula: ; in, i For depth indexing, j For distance index, z i For the first i The depth value corresponding to each depth grid point.

[0015] Step 10: Perform percentile color level cropping on the fusion matrix F; finally output the Mars shallow subsurface dual-scale fused radar profile; specifically in step 10, the lower limit is 1%-5%, and the upper limit is 95%-99%; use grayscale color matching to generate the depth domain radar profile, the vertical and horizontal size ratio of the image is in the range of 1:2.5-1:4, the output resolution is not less than 300 DPI, the vertical axis is labeled with depth, and the horizontal axis is labeled with distance, thus obtaining the Mars shallow subsurface dual-scale fused radar profile.

[0016] The technical effects achieved by this invention are as follows: Existing methods rely solely on single-radar imaging, failing to achieve both shallow resolution and deep penetration. This invention integrates RIMFAX and RoPeR dual-source data to achieve coordinated imaging with shallow precision and deep penetration over a depth range exceeding 100 meters.

[0017] Existing methods rely on commercial software or manual stitching, resulting in poor format compatibility. This invention eliminates stitching artifacts through grid alignment and Gaussian weighted transition, significantly improving processing efficiency and image quality. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall concept of the present invention. Figure 2 is a diagram of the original RIMFAX SEG-Y data and the original binary data file of RoPeR .2c of this invention; wherein, Figure 2a Yes: Screenshot of raw RIMFAX SEG-Y data; Figure 2b Yes: Screenshot of the raw binary data file of RoPeR .2c; Figure 3 This is the original cross-sectional view of the shallow RIMFAX layer of this invention. Figure 3 Showing striped noise and deep attenuation in the raw data; Figure 4 This is the original cross-sectional view of the deep RoPeR of this invention; Figure 4 Displaying low-resolution and weak signal characteristics of deep data; Figure 5 This is a cross-sectional view after shallow RIMFAX treatment according to the present invention; Figure 5 The filtered and enhanced shallow structure is clearly displayed; Figure 6 This is a cross-sectional view after deep RoPeR processing according to the present invention. Figure 6 Showing the deep layer contours after filtering enhancement; Figure 7 This is a cross-sectional view of the fusion radar, which is the core achievement of this invention. Figure 7 Show the complete cross-section after the light and dark areas are blended; Figure 8 This is a side-by-side comparison of the cross-section of this invention project and a typical cross-section from the literature. Detailed Implementation

[0019] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0020] like Figure 1 As shown, a dual-scale data fusion imaging method for Mars radar rover data includes the following steps: Step 1: Acquire RIMFAX shallow radar data and RoPeR deep radar data from the Mars rover; Step 2: Read RIMFAX data using the segyio library in Python, set ignore_geometry=True to bypass the incomplete gather definition, and set the original gather matrix as follows. ,in The number of time sampling points, The number of channels; the original sampling interval is Lane spacing is Uniform downsampling is performed in both the time and spatial domains; lateral sampling step size. Take 8-15, and extract the step size vertically. Take values ​​from 2 to 5, and then downsample the matrix. Defined as: ; in , The equivalent time sampling interval after downsampling becomes: The equivalent track spacing becomes To ensure that the signal does not alias, the highest frequency of the original signal should meet the following requirements. Extract the sampling interval dt; the horizontal sampling step size is 8-15, and the vertical sampling step size is 2-5. After downsampling the original trace gather, store it as a two-dimensional data matrix. A shallow ; Step 3: Read RoPeR .2c data file by file according to the preset sampling point count of 1024, and reconstruct the gather matrix; the horizontal trace extraction step size is 1-3, and the vertical sampling extraction step size is 1-3. After batch reading, the data is horizontally stitched together to generate a deep data matrix. A deep The process is as follows: in The total number of samples corresponding to the file size in bytes; preset number of sampling points per track. Then the first The number of channels in each file is Reconstruct into a matrix; then adjust by horizontal step size. Take 1-3 and longitudinal step size Take 1-3 downsampling samples: ; All K The files are concatenated horizontally to obtain a deep data matrix: ; Step 4: Perform shallow data filtering and enhancement processing; Step 4 specifically includes the following steps: Step 401: Perform DC removal: The DC component is caused by instrument bias or background noise, and its mathematical expectation is a constant; let the matrix... The elements are Total mean for: ; After removing DC: ; This operation normalizes the signal to zero, eliminates zero-frequency components, and prevents subsequent filtering from being affected by the bias. Step 402: Perform 2D Median Filtering: Median filtering is a nonlinear smoothing filter that is highly effective at suppressing striped outlier impulse noise. (Regarding window size...) (Odd number), define the neighborhood set: ; The filtered output is the median of all values ​​in the neighborhood: ; The mathematical basis of median filtering is order statistics, which have a zero impulse response, effectively preserving edges while filtering out isolated noise points. Shallow data is taken from... , .

[0021] Step 403: Perform moving average smoothing over time. A one-dimensional moving average is performed along the time axis, with a window length of [missing information]. (Odd number); Let the current sampling point be... Neighborhood range ,but: ; Mean smoothing is equivalent to convolution with a rectangular window; its frequency response is a sinc function, exhibiting low-pass filtering characteristics, and its cutoff frequency is... ; shallow layer ; Step 404: Perform automatic gain control based on Hilbert envelope (Hilbert Envelope AGC): Ground-penetrating radar (GPR) signals decay exponentially with depth; AGC compensates for energy loss through time-varying gain. First, the signal envelope is calculated. For the... channel signal Its analytical signal is: ; Hilbert transform Defined as: ; The Hilbert transform in the frequency domain corresponds to a phase shift. Positive frequency and Negative frequency; envelope is the modulus of the analytic signal: ; AGC in the sliding window Internal calculation of average envelope: ; The signal after gain compensation is: ; in It is a small positive number (such as This prevents division by zero; the operation normalizes the signal energy within a local time window, enhancing weak reflections in deeper areas. The shallow AGC window length is... .

[0022] Step 405: Perform destriping operation: Lateral stripe noise manifests as lateral inconsistency in the DC bias of each channel, and its mathematical model is additive stripe interference. Let the first... The horizontal mean of the time sampling points is: ; To remove the bands, simply subtract the mean: ; This process eliminates the difference in the transverse DC component, resulting in a more balanced grayscale distribution in the profile.

[0023] Step 406: Perform Gaussian smoothing operation: A two-dimensional Gaussian filter is a linear smoothing filter whose impulse response is a Gaussian function, exhibiting optimal time-frequency localization characteristics in both the spatial and frequency domains. The two-dimensional Gaussian kernel is defined as: ; in The longitudinal standard deviation controls the smoothness of the time direction. The lateral standard deviation controls the smoothness between channels. The smoothing operation is convolution: ; The frequency domain response of Gaussian filtering remains a Gaussian function, with no ringing effect, and the smoothed signal has good continuity. Shallow layer... , Final output .

[0024] Step 5: Perform deep data filtering and enhancement processing; Step 5 specifically includes the following steps: Step 501: Perform DC removal operation, the formula is the same as in step 401; Step 501: Perform two-dimensional median filtering: increase the window size to... , This is to address more severe transverse stripe interference. The mathematical definition is consistent with step 402; Step 503: Perform time-direction mean smoothing: window length This further suppresses random noise. The formula is the same as in step 403.

[0025] Step 504: Perform Hilbert envelope AGC operation: Increase AGC window length to This is to match the slower energy decay trend of deeper signals. The envelope and gain formulas are the same as in step 404; Step 505: Perform strip removal operation: Same as step 405; Step 506: Perform Gaussian smoothing: Compared to step 406, the smoothing intensity is increased. , This improves horizontal continuity. The formula is the same as in step 406; Step 507: Perform nonlinear amplitude compression (Logarithmic Compression): Deep signals have a large dynamic range, and strong reflections may mask weak reflections. Logarithmic compression is used to compress the amplitude range without changing the sign: ; This transformation originates from the Weber-Fechner law in psychology, which maps a large dynamic range to a more balanced interval. In the small amplitude region, it approximates linearity. The large value region is logarithmically compressed, thus highlighting the deep, weak in-phase axis.

[0026] Step 6: Perform time-depth transformation and coordinate generation; Specifically, in step 6: the time sampling interval is set to 0.8-1.2 ns, the shallow wave velocity to 0.08-0.12 m / ns, the deep wave velocity to 0.06-0.10 m / ns, and the channel spacing to 0.4-0.6 m; and the values ​​are calculated according to the formulas: Depth axis: Distance axis: ; Generate shallow depth axes respectively z shallow With depth axis z deep Distance axes are generated according to track number and track spacing respectively. x shallow andx deep .

[0027] Step 7: Perform unified grid control registration; Specifically, in step 7: Define a uniform depth grid number of 800-1500 and a uniform distance grid number of 500-1000. The depth range is the smaller of the maximum depths of the two sets of data, and the distance range is the smaller of the maximum distances of the two sets of data. Two-dimensional linear interpolation is used. ; First, interpolate the data along the depth direction to a unified depth axis, then interpolate it along the distance direction to a unified distance axis to obtain a spatially aligned matrix. B shallow and B deep .

[0028] Step 8: Define the Gaussian weighting function for the depth domain; Specifically, in step 8: Transition depth z c Constructing with depth coordinate z as independent variable Shallow weight function: ; Deep weight function: .

[0029] Step 9: Perform a point-by-point weighted summation of the fusion matrix F using the following formula: ; in, i For depth indexing, j For distance index, z i For the first i The depth value corresponding to each depth grid point.

[0030] Step 10: Perform percentile color level clipping on the fusion matrix F; finally output the Mars shallow subsurface dual-scale fused radar profile.

[0031] In step 10, the lower limit is set to 1%-5%, and the upper limit is set to 95%-99%. A depth domain radar profile is generated using grayscale color matching. The ratio of the vertical to horizontal dimensions of the image is 1:2.5-1:4, and the output resolution is not less than 300 DPI. The vertical axis is labeled with depth, and the horizontal axis is labeled with distance, thus obtaining a dual-scale fused radar profile of the shallow Martian subsurface.

[0032] This invention configures differentiated filtering parameters and gain strategies for RIMFAX shallow high-resolution data and RoPeR deep large-penetration data respectively, solving the problem that a single radar payload cannot simultaneously achieve both detection depth and imaging resolution. This invention achieves integrated imaging of the Martian shallow substructure from the surface to the deep contours by using unified grid resampling and depth domain Gaussian weighted transition fusion to accurately register and smoothly connect dual-source heterogeneous radar data in space.

[0033] The working principle of this invention is as follows: As shown in step 1, obtain RIMFAX shallow radar SEG-Y format data and RoPeR deep radar binary .2c format data, as shown in Figure 2. As shown in step 2, read the RIMFAX data, extract the sampling interval, downsample with a horizontal step size of 10 and a vertical step size of 3 to generate a shallow data matrix, as shown in the attached figure. Figure 3 As shown; Figure 3 Showing striped noise and deep attenuation in the raw data; As shown in step 3, the .2c data is reconstructed with 1024 sampling points per channel, downsampled with a horizontal step size of 2 and a vertical step size of 2, and batch-stitched to generate a deep data matrix, as shown in the attached figure. Figure 4 As shown; Figure 4 Displaying low-resolution and weak signal characteristics of deep data; As shown in steps 4 and 5, the shallow matrix is ​​sequentially subjected to DC removal, two-dimensional median filtering, time smoothing, Hilbert envelope AGC, striping removal, and Gaussian smoothing. The processed shallow profile is shown in the attached figure. Figure 5 As shown; Figure 5 The image shows a clearer shallow structure after filtering enhancement; the same process is performed on the deep matrix with the addition of nonlinear amplitude compression, and the processed deep profile is shown in the attached image. Figure 6 As shown; Figure 6 Showing the deep layer contours after filtering enhancement; Based on steps 6 and 7, the depth axis and distance axis of the two sets of data are calculated respectively, a unified grid is defined and resampling is performed through two-dimensional linear interpolation to obtain spatially aligned shallow and deep matrices. Based on steps 8 and 9, the transition depth is set, and a Gaussian weighting function is constructed to weight and fuse the registered matrices. As shown in step 10, the fusion matrix is ​​cropped to percentile color levels and visualized in grayscale, ultimately outputting a shallow radar profile of Mars that combines shallow fine structure with deep penetration information, as shown in the attached figure. Figure 7 As shown; Figure 7 Show the complete cross-section after the light and dark areas are blended; The merged profile was compared with typical RIMFAX and RoPeR profiles in published literature, as shown in the appendix. Figure 8 As shown, Figure 8 This is a juxtaposed comparison of the fusion profile of this invention and a typical profile from the literature; it can be seen that the layered structure and reflection mode are highly consistent, but this invention achieves a dual-scale collaborative imaging effect of shallow, fine and deep.

[0034] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A dual-scale data fusion imaging method for Mars radar rover data, characterized in that: Includes the following steps: Step 1: Acquire RIMFAX shallow radar data and RoPeR deep radar data from the Mars rover; Step 2: Read RIMFAX data using the segyio library in Python, set ignore_geometry=True to bypass the incomplete gather definition, and set the original gather matrix as follows. ,in The number of time sampling points, The number of channels; the original sampling interval is Lane spacing is Uniform downsampling is performed in both the time and spatial domains; lateral sampling step size. Take 8-15, and extract the step size vertically. Take values ​​from 2 to 5, and then downsample the matrix. Defined as: ； in , The equivalent time sampling interval after downsampling becomes: The equivalent track spacing becomes The highest frequency of the original signal should meet the following requirements. ; Extract the sampling interval dt; downsample the original gather and store it as a two-dimensional data matrix. A shallow ; Step 3: Read RoPeR .2c data file by file according to the preset sampling point count of 1024, and reconstruct the gather matrix; the horizontal trace extraction step size is 1-3, and the vertical sampling extraction step size is 1-3. After batch reading, the data is horizontally stitched together to generate a deep data matrix. A deep The process is as follows: in The total number of samples corresponding to the file size in bytes; preset number of sampling points per track. Then the first The number of channels in each file is Reconstruct into a matrix; then adjust by horizontal step size. Take 1-3 and longitudinal step size Take 1-3 downsampling samples: ； All K The files are concatenated horizontally to obtain a deep data matrix: ; Step 4: Perform shallow data filtering and enhancement processing; Step 5: Perform deep data filtering and enhancement processing; Step 6: Perform time-depth transformation and coordinate generation; Step 7: Perform unified grid control registration; Step 8: Define the Gaussian weighting function for the depth domain; Step 9: Perform point-by-point weighted summation of the fusion matrix F using the following formula; Step 10: Perform percentile color level clipping on the fusion matrix F; finally output the Mars shallow subsurface dual-scale fused radar profile.

2. The dual-scale data fusion imaging method for Mars rover data according to claim 1, characterized in that: Step 4 specifically includes the following steps: Step 401: Perform DC removal operation: Let matrix The elements are Total mean for: ； After removing DC: ； Step 402: Perform 2D Median Filtering: For window size For odd numbers, define the neighborhood set: ； The filtered output is the median of all values ​​in the neighborhood: ； Shallow data retrieval , ; Step 403: Perform time-direction mean smoothing: A one-dimensional moving average is performed along the time axis, with a window length of [missing information]. (Odd number); Let the current sampling point be... Neighborhood range ,but: ； Mean smoothing is equivalent to convolution with a rectangular window; its frequency response is a sinc function, exhibiting low-pass filtering characteristics, and its cutoff frequency is... ; shallow layer ; Step 404: Perform automatic gain control based on Hilbert envelope: First, calculate the signal envelope; for the... channel signal Its analytical signal is: ； Hilbert transform Defined as: ； Where pv is the Cauchy principal value; It is an integral variable; For dummy variables in Hilbert integration; for phase shifts in the Hilbert transform in the frequency domain. Positive frequency and Negative frequency; envelope is the modulus of the analytic signal: ； Automatic gain control (AGC) in a sliding window Internal calculation of average envelope: ； The signal after gain compensation is: ； in It is a small positive number; this operation normalizes the signal energy within a local time window, and the shallow AGC window length is taken as... ; Step 405: Perform strip removal operation: Let the first The horizontal mean of the time sampling points is: ； To remove the bands, subtract the mean: ； Step 406: Perform Gaussian smoothing: The two-dimensional Gaussian kernel is defined as: ； in The longitudinal standard deviation controls the smoothness of the time direction. The lateral standard deviation controls the smoothness between channels; the smoothing operation is convolution. ； shallow layer , Final output .

3. The dual-scale data fusion imaging method for Mars radar rover data according to claim 2, characterized in that: Step 5 specifically includes the following steps: Step 501: Perform DC removal operation, the formula is the same as in step 401; Step 501: Perform two-dimensional median filtering: increase the window size to... , The mathematical definition is consistent with step 402. Step 503: Perform time-direction mean smoothing: window length The formula is the same as in step 403; Step 504: Perform Hilbert envelope AGC operation: Increase AGC window length to The envelope and gain formulas are the same as in step 404; Step 505: Perform strip removal operation: Same as step 405; Step 506: Perform Gaussian smoothing: Compared to step 406, the smoothing intensity is increased. , The formula is the same as in step 406; Step 507: Perform nonlinear amplitude compression: Use logarithmic compression to compress the amplitude range without changing the sign. 。 4. The dual-scale data fusion imaging method for Mars radar rover data according to claim 3, characterized in that: Specifically, in step 6: the time sampling interval is set to 0.8-1.2 ns, the shallow wave velocity to 0.08-0.12 m / ns, the deep wave velocity to 0.06-0.10 m / ns, and the channel spacing to 0.4-0.6 m; and the values ​​are calculated according to the formulas: Depth axis: Distance axis: ; Generate shallow depth axes respectively z shallow With depth axis z deep Distance axes are generated according to track number and track spacing respectively. x shallow and x deep .

5. A dual-scale data fusion imaging method for Mars rover data according to claim 4, characterized in that: Specifically, in step 7: Define a uniform depth grid number of 800-1500 and a uniform distance grid number of 500-1000. The depth range is the smaller of the maximum depths of the two sets of data, and the distance range is the smaller of the maximum distances of the two sets of data. Two-dimensional linear interpolation is used. ； Where, interp1d: represents "one-dimensional linear interpolation" operation, which infers the value of intermediate points by connecting known points with a straight line; U: the original radar data matrix to be interpolated; V: the intermediate transition matrix generated after the first step of interpolation; z old : The original depth coordinate sequence before interpolation; x old : The original distance coordinate sequence before interpolation; axis=0 specifies vertical interpolation along the depth axis of the matrix, i.e., fixing the horizontal distance; axis=1 specifies horizontal interpolation along the distance axis of the matrix, i.e., fixing the depth; First, the data is interpolated along the depth direction to a unified depth axis, and then interpolated along the distance direction to a unified distance axis, resulting in a spatially aligned matrix. B shallow and B deep .

6. The dual-scale data fusion imaging method for Mars radar rover data according to claim 5, characterized in that: Specifically, in step 8: Transition depth z c Constructing with depth coordinate z as independent variable Shallow weight function: ； Deep weight function: 。 7. The dual-scale data fusion imaging method for Mars radar rover data according to claim 6, characterized in that: The fusion matrix F in step 9 is calculated by weighted summation point by point according to the following formula: ； in, i For depth indexing, j For distance index, z i For the first i The depth value corresponding to each depth grid point.

8. The dual-scale data fusion imaging method for Mars radar rover data according to claim 1, characterized in that: In step 10, the lower limit is 1%-5% and the upper limit is 95%-99%. A depth domain radar profile is generated using grayscale color matching. The ratio of the vertical to horizontal dimensions of the image is 1:2.5-1:4, and the output resolution is not less than 300 DPI. The vertical axis is labeled with depth and the horizontal axis is labeled with distance, thus obtaining a dual-scale fused radar profile of the shallow subsurface of Mars.

Images