Orthogonal gpr fracture enhanced imaging method based on structural confidence constraint

Abstract

The application relates to the technical field of ground penetrating radar (GPR) data processing, and discloses a structure confidence constraint-based orthogonal GPR fissure enhanced imaging method. The method comprises the following steps: acquiring X and Y direction GPR data in the same detection area, and generating C-scan images respectively; taking the X direction as a reference to unify the space coordinates and perform physical grid resampling on the Y direction; performing double-layer decomposition on the two groups of images to obtain a basic layer and a detail layer; calculating a motion degree map based on convolution sparse representation and structure confidence based on a local structure tensor for the detail layer; constructing a detail reliability score, and adaptively determining a fusion weight to obtain a fused detail layer; performing information entropy weight fusion on the basic layer to obtain a fused basic layer; and reconstructing the fused basic layer and the fused detail layer to obtain a final fissure enhanced image. The application can comprehensively integrate X / Y direction fissure complementary responses, effectively suppress background clutter and isolated noise interference, and improve the continuity, visibility and spatial correspondence accuracy of the fissure structure.

Description

Technical Field

[0001] This invention relates to the field of ground-penetrating radar data processing technology, and in particular to an orthogonal GPR crack enhancement imaging method based on structural confidence constraints. Background Technology

[0002] Ground penetrating radar (GPR) is a non-destructive testing technology that uses electromagnetic waves to detect underground structures. It offers advantages such as high detection speed, strong continuous imaging capability, and applicability to shallow underground defects. In underground fracture detection, the spatial distribution of fractures or anomalies is often analyzed using B-scan and C-scan image formats. C-scan images (also known as depth slices or time slices) are planar response images obtained by projecting sampling points within the same depth (or time) range from 3D GPR data along the horizontal direction, providing a direct visual representation of the distribution of underground targets on the horizontal plane.

[0003] Existing ground-penetrating radar C-scan imaging methods typically rely on single-direction scan data, generating planar response images through time-window averaging, channel interpolation, or simple energy superposition. While these methods are simple to implement, they suffer from the following problems: First, the results of single-direction scanning are significantly affected by the survey line direction. Studies have shown that the survey line direction of the ground-penetrating radar profile has a significant impact on the detection results. When the fracture direction does not match the survey line direction or the antenna polarization direction, the fracture response is prone to discontinuities or missing information. For example, when the survey line direction is parallel to the linear target orientation, it is difficult to obtain an effective reflection response. Although existing technologies have schemes for acquiring three-dimensional ground-penetrating radar data by performing grid scanning along two orthogonal directions, a systematic fusion enhancement method targeting the complementary features of the fracture direction has not yet been formed.

[0004] Second, simple averaging, local energy weighting, or image overlay methods are easily affected by strong background reflections, isolated noise, and local high-energy artifacts, which may weaken the true fracture details or enhance non-target structures. Since ground-penetrating radar C-scan images contain a large amount of background clutter and noise interference, it is difficult to distinguish between the true fracture response and noise artifacts by fusing based solely on amplitude or energy magnitude.

[0005] Third, existing multi-frequency or multi-source ground-penetrating radar fusion methods mostly target frequency complementarity or multi-sensor complementarity, with fewer methods addressing the fusion enhancement of complementary features in the crack direction of X / Y orthogonal scan data. Methods such as gradient structure tensor have seen preliminary applications in ground-penetrating radar data analysis, but have not yet been used for structural reliability assessment and adaptive weight allocation in orthogonal C-scan image fusion.

[0006] Therefore, a crack enhancement imaging method for orthogonal ground-penetrating radar data needs to be proposed. Based on the unification of X / Y orthogonal data spatial coordinates, the crack response under different scanning directions is integrated, and the influence of background texture and isolated noise on the fusion result is suppressed, thereby improving the visibility, continuity and spatial correspondence of crack structures in C-scan images. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides an orthogonal GPR crack enhancement imaging method based on structural confidence constraints, aiming to comprehensively utilize the complementary features of crack directions in X / Y orthogonal scanning data, suppress background clutter and isolated noise interference, and improve the continuity, visibility, and spatial correspondence accuracy of crack structures in C-scan images.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows: An orthogonal GPR crack enhancement imaging method based on structural confidence constraints includes the following steps: Step 1: Acquire ground-penetrating radar data in the X and Y directions within the same detection area, and generate corresponding X and Y direction C-scan images respectively, wherein the X and Y directions are orthogonal to each other; Step 2: Using the C-scan image in the X direction as a reference, perform spatial coordinate unification and physical grid resampling on the C-scan image in the Y direction, so that the two sets of C-scan images are in the same spatial coordinate system and under the same physical grid. Step 3: Perform two-layer decomposition on the C-scan images in the X and Y directions after unifying the spatial coordinates to obtain their respective base layer and detail layer; Step 4: Perform directional filtering and sparsification based on convolutional sparse representation on the detail layers in the X and Y directions respectively to obtain the activity maps of the detail layers in the X and Y directions; at the same time, calculate the structure confidence of the detail layers in the X and Y directions respectively. Step 5: Construct detail reliability scores for the X and Y directions by combining the activity maps of the detail layers in the X and Y directions with the corresponding structural confidence scores. Adaptively determine the fusion weights of the detail layers in the X and Y directions based on the detail reliability scores in the X and Y directions to obtain the fused detail layers. Step 6: Perform information entropy weight fusion on the base layers in the X and Y directions to obtain the fused base layer; Step 7: Perform image reconstruction on the fused base layer and the fused detail layer to obtain the final crack-enhanced C-scan imaging result.

[0009] In the above scheme, step 1, generating C-scan images in the X and Y directions, specifically includes: Step 11, let the X-direction ground-penetrating radar data be... Y-direction ground-penetrating radar data is ,in, Indicates the time sampling point. and These represent two horizontal spatial directions in the same physical coordinate system. The current C-scan imaging time window is denoted as... ; Step 12: Perform local Gaussian averaging on the original ground-penetrating radar data to remove the background, and obtain the background-removed radar data. Step 13: Perform Kirchhoff migration imaging processing on the background-removed ground-penetrating radar data; Step 14, after Kirchhoff offset processing, within the selected time window Extract C-scan images in the X direction separately. and C-scan image in the Y direction .

[0010] In the above scheme, step 2, which involves unifying the spatial coordinates and resampling the physical grid of the C-scan image in the Y direction, specifically includes: Step 21: Transpose the C-scan image in the Y direction so that its matrix axis is consistent with the C-scan image in the X direction; Step 22: Perform flipping and translation correction on the transposed Y-direction C-scan image to obtain a spatially aligned Y-direction image; Step 23: Establish a unified physical grid based on the actual sampling interval of the X and Y direction data, and resample the two sets of C-scan images to the same spatial scale to obtain the resampled X-direction C-scan image. and C-scan image in the Y direction .

[0011] In the above scheme, step 3 uses the Tikhonov decomposition method for the two-layer decomposition. Taking the X-direction image as an example, the base layer is obtained through frequency domain low-pass filtering, and the detail layer is obtained by subtracting the base layer from the original image, i.e.: ; ; in, and These represent the base layers of the images in the X and Y directions, respectively, and contain low-frequency background, overall energy distribution, and gradually varying regions. and These represent detail layers of the images in the X and Y directions, respectively, and include crack edges, local mutations, high-frequency textures, and directional structural responses.

[0012] In the above scheme, step 4 involves performing directional filtering and sparsification based on convolutional sparse representation on the detail layers in the X and Y directions to obtain activity maps of the detail layers in the X and Y directions. Specifically, this includes: Step 41: Perform convolution filtering on the X-direction detail layer and the Y-direction detail layer respectively using a directional filter bank to obtain the corresponding filter response; Step 42: Perform soft threshold sparsification on the filter response in each direction to obtain a sparse coefficient map; Step 43: Calculate the activity map of the detail layer in the X direction based on the sparse coefficient map. Activity plot of detail layer in the Y direction .

[0013] In the above scheme, step 4, calculating the structural confidence of the detail layers in the X and Y directions, specifically includes: Step 44, for the detail layer in any direction Calculate its horizontal and vertical gradients: ; Step 45: Construct the three components of the structure tensor based on the local gradient: ; ; ; in, Use a Gaussian smoothed window. This represents the convolution operation. , , These are the three components of the local structure tensor, used to describe the gradient energy and orientation distribution characteristics of the local structure; Step 46: Calculate structural orientation consistency based on the structural tensor: ; in, To ensure structural consistency, To prevent extremely small positive numbers with a denominator of zero; Step 47, calculate the local structural energy: ; in, For local structural energy, This indicates normalization processing; Step 48: Combine structural energy and structural orientation consistency to obtain structural confidence level: ; in, For structural confidence, and These are the weighting coefficients; For detail layers in the X direction respectively and Y-direction detail layer Perform steps 44 to 48 to obtain the structural confidence of the detail layer in the X direction. and the structural confidence of the detail layer in the Y direction .

[0014] In the above scheme, step 5 specifically includes: Step 51: Combine the activity map with the structural confidence score to construct a detailed reliability score. ; ; in, , The reliability scores for the detail layers in the X and Y directions are respectively. , These are the activity maps for the detail layers in the X and Y directions, respectively. , These represent the structural confidence levels of the detail layers in the X and Y directions, respectively. and This is the weighting adjustment coefficient. This indicates normalization processing; Step 52: Construct the fusion weights for the X-direction detail layer and the Y-direction detail layer based on the detail reliability score. ; ; in, , These are the fusion weights for the detail layers in the X and Y directions, respectively. As a weighting enhancement factor, To prevent extremely small positive numbers with a denominator of zero; Step 53, obtaining the fusion detail layer under structural confidence constraints: ; in, To integrate the detail layer, , These are detail layers for the X and Y directions, respectively.

[0015] In the above scheme, step 6 specifically includes: Step 61, calculate the base layer in the X direction respectively. and Y-direction base layer Information entropy and ; Step 62: Determine the base layer fusion weights based on information entropy: ; ; in, , These are the fusion weights of the base layers in the X and Y directions, respectively. , These are the information entropies of the base layer in the X and Y directions, respectively. To prevent extremely small positive numbers with a denominator of zero; Step 63: Weight the X-direction base layer and the Y-direction base layer according to the base layer fusion weights to obtain the fused base layer:

[0016] in, To integrate the base layer.

[0017] In the above scheme, step 7 involves adding the fusion base layer and the fusion detail layer to obtain the final crack-enhanced C-scan imaging result:

[0018] in, For the final crack-enhanced C-scan image, To integrate the base layer, To integrate the detail layer, This is the detail gain coefficient, used to adjust the enhancement level of the crack detail layer in the final image.

[0019] An orthogonal GPR crack enhancement imaging system based on structural confidence constraints, comprising: The data acquisition module is used to acquire X-direction ground-penetrating radar data and Y-direction ground-penetrating radar data within the same detection area, and generate corresponding X-direction C-scan images and Y-direction C-scan images respectively, wherein the X-direction and the Y-direction are orthogonal to each other; The spatial registration module is used to unify the spatial coordinates and resample the physical grid of the C-scan image in the Y direction with the X-direction C-scan image as a reference, so that the two sets of C-scan images are in the same spatial coordinate system and under the same physical grid. The two-layer decomposition module is used to perform two-layer decomposition on the X-direction C-scan image and the Y-direction C-scan image after the spatial coordinates are unified, so as to obtain their respective base layer and detail layer; The structure confidence calculation module is used to calculate the activity maps based on convolutional sparse representation for the X-direction detail layer and the Y-direction detail layer respectively, and to calculate the structure confidence based on the local structure tensor respectively. The detail layer fusion module is used to jointly construct a detail reliability score by combining the activity map with the corresponding structural confidence score, and adaptively determine the fusion weight based on the detail reliability scores in the X and Y directions to obtain the fused detail layer. The base layer fusion module is used to perform information entropy weight fusion on the X-direction base layer and the Y-direction base layer to obtain a fused base layer. The image reconstruction module is used to reconstruct the image of the fused base layer and the fused detail layer to obtain the final crack-enhanced C-scan imaging result.

[0020] Through the above technical solution, the orthogonal GPR crack enhancement imaging method based on structural confidence constraints provided by the present invention has the following beneficial effects: First, this invention utilizes the complementarity of orthogonal ground-penetrating radar data in the X and Y directions to synthesize the fracture response under different scanning directions, thus improving the problems of incomplete fracture response and strong direction dependence in single-direction detection. By fusing orthogonal scanning data in the X and Y directions, the sensitive response of each direction to fractures with specific orientations is preserved, thereby improving the integrity and spatial continuity of fracture imaging. Secondly, this invention improves the spatial correspondence accuracy of the fusion result by unifying spatial coordinates and resampling the physical grid, enabling the fusion of two sets of orthogonal C-scan images in the same physical coordinate system. Specifically, by transposing, flipping, and translating the Y-direction C-scan image and combining it with interpolation resampling using a unified physical grid, the problem of the inability to directly fuse X-direction and Y-direction data due to differences in acquisition direction, matrix axis, and sampling interval is solved. Third, this invention decomposes C-scan images into a base layer and a detail layer, and uses different strategies to fuse them, preserving the overall energy distribution while enhancing the response of crack edges and high-frequency structures. Base layer fusion preserves the low-frequency background and gradually varying region information of the image, while detail layer fusion focuses on high-frequency features such as crack edges, local abrupt changes, and directional structures, achieving hierarchical and targeted fusion enhancement. Fourth, this invention proposes a detail-level fusion strategy constrained by structural confidence. By calculating structural orientation consistency and structural energy using local structural tensors, it can determine whether local details possess continuous, well-defined crack structures, thereby suppressing noise and unreliable strong responses. Structural confidence simultaneously considers the gradient energy and orientation distribution characteristics of the local structure, suppressing noise responses with large amplitudes but chaotic orientations, and enhancing crack responses with moderate amplitudes but continuous orientations and obvious linear characteristics, effectively reducing the interference of background clutter and isolated noise on the fusion results. Fifth, this invention jointly constructs a detail reliability score by combining the activity of convolutional sparse representation (CSR) and structural confidence, and adaptively allocates fusion weights for detail layers in the X and Y directions accordingly, making the crack structure in the fused image more continuous, clear, and reliable. Only regions with both high sparse detail intensity and high structural reliability can obtain higher weights in the fusion, avoiding the pseudo-detail responses introduced by traditional methods that simply fuse based on amplitude strength. Sixth, the method of this invention is applicable to enhanced imaging scenarios of multi-channel ground-penetrating radar, orthogonal ground-penetrating radar, and underground anomalies such as fissures, cavities, and voids, and has good engineering application value. This method can be integrated into ground-penetrating radar data processing software, providing more reliable imaging data for refined detection of underground fissures and engineering evaluation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0022] Figure 1 This is a schematic diagram of the process of an orthogonal GPR crack enhancement imaging method based on structural confidence constraints disclosed in an embodiment of the present invention; Figure 2 The images are C-scan images of crack enhancement processed by different fusion methods, where (a) is average fusion, (b) is wavelet transform fusion, (c) is local energy weighted fusion, and (d) is the CSR fusion based on structural confidence constraints of this invention. Figure 3 This is a schematic diagram of an orthogonal GPR crack enhancement imaging system based on structural confidence constraints disclosed in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0024] I. Method Implementation Examples This invention provides an orthogonal GPR crack enhancement imaging method based on structural confidence constraints, such as... Figure 1 As shown, the specific implementation method is illustrated using orthogonal ground-penetrating radar data in the X and Y directions within the same detection area as an example. Data acquisition can be accomplished using a multi-channel ground-penetrating radar system or by acquiring data in stages using a single-channel ground-penetrating radar; this invention does not limit the method. In the following implementation method, a multi-channel ground-penetrating radar system is used as an example. This system includes a transmitting antenna, a receiving antenna, a control unit, and a data acquisition unit. The X-direction and Y-direction survey lines are arranged in an orthogonal grid within the same detection area, and the spacing between the survey lines and the sampling interval are determined according to the detection depth and resolution requirements.

[0025] Step 1: Data Acquisition and C-scan Image Generation Acquire ground-penetrating radar data in the X and Y directions within the same detection area, and generate corresponding X-direction C-scan images respectively. and C-scan image in the Y direction The X and Y directions are orthogonal to each other.

[0026] Step 11, let the X-direction ground-penetrating radar data be... Y-direction ground-penetrating radar data is ; in, Indicates the time sampling point, in ns; and These represent two horizontal spatial directions in a unified physical coordinate system, with units of meters (m). The current C-scan imaging time window is denoted as... , The start time, The time interval is the end time, and the unit is ns.

[0027] Step 12: Perform local Gaussian averaging on the original ground-penetrating radar data to remove the background, and obtain the background-removed radar data.

[0028] The background component is estimated along each survey line direction by using a local Gaussian kernel, and the background component is subtracted from the original data to suppress continuous background response, direct wave interference, and systematic strip noise.

[0029] For radar data in any direction The background removal result is expressed as: ; in, This represents the radar data after background removal; The background component obtained by local Gaussian mean estimation is calculated using the following formula: ; ; In the formula, The radius of the local window is expressed in units of the number of sampling points. The standard deviation of the Gaussian kernel controls the smoothness of the Gaussian kernel; To normalize the Gaussian weights, satisfying ; These are the sampled values ​​of radar data within the spatial neighborhood. The background removal processing described above is performed along each data track, i.e., for a fixed... and ,along The direction is smoothed using one-dimensional Gaussian smoothing, and the resulting background estimate is subtracted from the original data. In this embodiment, The value of is determined based on the spatial scale of background response in the actual data, with a typical value being . Preferred ; The value is usually .

[0030] Step 13: Perform Kirchhoff migration imaging processing on the background-removed ground-penetrating radar data.

[0031] Kirchhoff offset imaging processing is used to enhance the spatial focusing effect of the actual reflection response and reduce the influence of the hyperbolic diffusion response on the C-scan image.

[0032] Let the current imaging point be ,in, For the two-way travel time The equivalent depth obtained from the conversion: , The unit is meters. The speed of electromagnetic waves in a medium is expressed in m / ns. This represents the two-way propagation time, measured in nanoseconds (ns). It applies to neighboring sampling points within the aperture range. The corresponding two-way propagation time is: ; in, , , and These are the spatial sampling intervals in the X and Y directions, respectively, in meters. and These are offset indices within the spatial aperture, all of which are integers.

[0033] The Kirchhoff migration imaging results can be expressed as: ; in, This is the result of offset imaging; and These represent the offset aperture radii in the X and Y directions, respectively, in units of the number of sampling points; These are the aperture weighting factors; To prevent extremely small constants with a denominator of zero, a typical value is taken as... .

[0034] Aperture weight It can be written as: ; The tilt angle weight is: ;

[0035] Hann aperture window is: ;

[0036] The distance compensation weight is: ; in, , representing the horizontal distance from the current sampling point to the imaging point, in meters; if distance compensation is not enabled, then let Tilt angle weight Hann aperture window used to compensate for geometric diffusion effects at tilted reflective interfaces Distance compensation weights are used to suppress the cutoff effect at the aperture edge. Used to compensate for energy loss that occurs with increasing distance.

[0037] Step 14, after Kirchhoff offset processing, within the selected time window Extract C-scan images in the X direction separately. and C-scan image in the Y direction .

[0038] within the selected time window Extract C-scan images in the X and Y directions respectively: ; ; in, Indicates time window The number of sampling points within; Represents the C-scan image in the X direction; Represents the C-scan image in the Y direction; and These are the Kirchhoff offset results for the X and Y directions, respectively. (Time window) The value is determined based on the burial depth range of the target fracture, and the specific value is determined by the electromagnetic parameters of the target and the medium.

[0039] Step 2: Spatial coordinate unification and physical grid resampling C-scan image in the X direction For reference, the C-scan image in the Y direction Spatial coordinate unification and physical grid resampling are performed to place the two sets of C-scan images in the same spatial coordinate system and under the same physical grid.

[0040] Since the scanning direction, matrix axis, and sampling interval of the ground-penetrating radar data in the X and Y directions may be different, it is necessary to first convert the two sets of C-scan images to the same spatial coordinate system and unified physical grid.

[0041] Step 21: Transpose the C-scan image in the Y direction so that its matrix axis is consistent with the C-scan image in the X direction.

[0042] First, transpose the C-scan image in the Y direction so that its matrix axis is aligned with the C-scan image in the X direction: ; in, This indicates that a matrix transpose operation is performed on the C-scan image in the Y direction. After transpose, the spatial coordinates of the image are changed from... Become .

[0043] Step 22: Flip and correct the position of the transposed Y-direction C-scan image to obtain a spatially aligned Y-direction image. .

[0044] Based on the actual acquisition starting point, scanning direction, and channel arrangement, the transposed Y-direction C-scan image is flipped vertically or horizontally, and the translation parameters are adjusted accordingly. and Perform position correction to obtain the spatially aligned Y-direction image: ; in, This indicates a spatial correction transformation that includes flipping and translation; and These are the translation parameters in the X and Y directions, respectively, in units of the number of sampling points; This represents the C-scan image in the Y direction, which is in the same spatial coordinate system as the X-direction image. It includes the flip type (vertical or horizontal) and translation parameters. , The coordinate origin, scanning start point, and channel arrangement relationship during actual data acquisition are determined, and are usually obtained through on-site measurement records or position information in the data header file.

[0045] Step 23: Establish a unified physical grid based on the actual sampling interval of the X and Y direction data, and resample the two sets of C-scan images to the same spatial scale to obtain the resampled X-direction C-scan image. and C-scan image in the Y direction .

[0046] Based on the actual trace spacing and channel spacing of the X and Y direction data, a unified physical grid is established, and the two sets of C-scan images are resampled to the same spatial scale. Let the query point on the unified physical grid be... The interpolated C-scan images in the X and Y directions are as follows: ; .

[0047] in, The interpolation operator can be either Linear interpolation or PCHIP interpolation (piecewise cubic Hermite interpolation polynomial). Linear interpolation is computationally efficient and suitable for scenarios with large amounts of data; PCHIP interpolation can better preserve the edge features of the image and is suitable for scenarios with complex crack structures. In this embodiment, PCHIP interpolation is preferred to preserve the boundary information of the crack as much as possible during the resampling process. Through the above processing, two C-scan images, X and Y, with consistent spatial coordinates and physical scale can be obtained. The resampled X-direction and Y-direction C-scan images are denoted as follows: , This provides a unified input for subsequent two-layer decomposition and fusion.

[0048] Step 3: Two-layer decomposition C-scan image in the X direction after spatial coordinate unification and C-scan image in the Y direction Two-layer decomposition is performed to obtain the base layer and detail layer respectively. The base layer mainly includes low-frequency background, overall energy distribution and gradually varying regions, while the detail layer mainly includes crack edges, local abrupt changes, high-frequency textures and directional structural responses.

[0049] The resampled C-scan images in the X and Y directions are denoted as follows: , .

[0050] Perform a two-layer decomposition on each image, breaking them down into a base layer and a detail layer: ; ; in, and These represent the base layers of the images in the X and Y directions, respectively, and mainly include low-frequency background, overall energy distribution, and gradually varying regions. and These represent detail layers of the image in the X and Y directions, respectively, and mainly include crack edges, local mutations, high-frequency textures, and directional structural responses.

[0051] In this embodiment, Tikhonov decomposition is used to obtain the base layer and detail layer. Taking the X-axis image as an example, its base layer can be represented as: ; ; The Y-direction image can be represented similarly. Wherein, and These represent the Fourier transform and the inverse Fourier transform, respectively. This represents the smoothing constraint parameter, which controls the smoothness of the base layer. The larger the value, the smoother the base layer and the fewer high-frequency components the detail layer contains. In this implementation, a typical value is [value to be filled in]. Preferred ; and This represents the discrete difference operator response in the frequency domain corresponding to the spatial directions, corresponding to the spatial frequencies in the X and Y directions, respectively. Tikhonov decomposition obtains a smooth base layer by suppressing high-frequency components in the frequency domain. Its physical meaning lies in decomposing an image into slowly changing background parts and rapidly changing details.

[0052] Step 4: Fusion of Convolutional Sparse Representations with Structural Confidence Constraints For the detail layer in the X direction and Y-direction detail layer By performing directional filtering and sparsification based on convolutional sparse representation, the activity map of the detail layer in the X direction is obtained. Activity plot of detail layer in the Y direction Simultaneously, the structural confidence of the detail layers in the X direction is calculated. and the structural confidence of the detail layer in the Y direction .

[0053] This step is the core of this invention. Traditional image fusion methods typically determine fusion weights based on local amplitude, gradient intensity, or activity level. However, in ground-penetrating radar (GPR) fracture imaging, strong local responses do not necessarily correspond to actual fracture structures; they may originate from noise, background remnants, isolated scattering points, or non-fracture targets. Therefore, this invention introduces structural confidence based on the sparse representation of activity through convolution to determine whether the detail response has continuity and clear directionality, thereby improving the reliability of fracture detail fusion.

[0054] 4.1 Calculation of Activity Degree in Convolutional Sparse Representation Step 41: Use directional filter banks to process the detail layers in the X direction. and Y-direction detail layer Perform convolution filtering to obtain a multi-directional filtered response.

[0055] Let the directional filter bank be ,in, Indicates the first One directional filter, Indicates the number of filters. A directional filter bank is used to extract structural features in different directions. The value of determines the directional resolution; in this embodiment, is taken as . That is, 8 equally spaced directions (0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135°, 157.5°). The value of can range from 4 to 16. The larger the dimension, the higher the resolution, but the computational load also increases accordingly.

[0056] Convolutional filtering is performed on the detail layers in the X and Y directions respectively to obtain the corresponding filtered responses: ; ; in, This represents the convolution operation; and These represent the detail layers in the X and Y directions, respectively. The filtering response under each directional filter.

[0057] Step 42: Perform soft threshold sparsification on the filter response in each direction to obtain a sparse coefficient map.

[0058] The filter response is subjected to soft-threshold sparsification to obtain a sparse coefficient map: ; ; in, and These represent the X and Y directions at the [number]th [year]. Sparse coefficients under directional filters; Indicates the sparse threshold; It is a symbolic function; This indicates that the value within the parentheses is the larger of the range between 0 and 0, meaning the positive value is retained. Soft thresholding will handle values ​​smaller than the threshold. The response is set to zero, and responses with amplitudes greater than the threshold are retained and the threshold is subtracted, thereby achieving sparsity of the coefficients.

[0059] In one specific implementation, the sparsity threshold can be expressed as: ; in, This is the threshold adjustment coefficient, and in this embodiment, it typically takes the value of [value missing]. Preferred ; This indicates median operation; 1.4826 is a constant factor used to convert the absolute deviation of the median into an unbiased estimate of the standard deviation. For the current detail layer ( or This threshold selection method can adaptively determine the sparse threshold based on the noise level of the data, and has good robustness.

[0060] Step 43: Calculate the activity map of the detail layer in the X direction based on the sparse coefficient map. Activity plot of detail layer in the Y direction .

[0061] L1 activity maps in the X and Y directions are calculated based on the sparse coefficient map to characterize the intensity of local detail response: ; ; in, This represents a local smoothing window or a Gaussian weighted window; and These represent the activity maps of the detail layers in the X and Y directions, respectively. This indicates the operation of taking the absolute value. The activity map is obtained by summing the absolute values ​​of the sparse coefficients in each direction and then smoothing them locally. The greater the activity, the stronger the local high-frequency detail response at that location.

[0062] 4.2 Calculation of structural confidence based on structural tensor Step 44, for the detail layer in any direction Calculate its horizontal and vertical gradients.

[0063] Structure confidence is calculated based on the local structure tensor. For detail layers in any direction... Calculate its horizontal and vertical gradients: ; ; in, and These are the first-order partial derivatives of the detail layer in the X and Y directions, i.e., the gradients in the horizontal and vertical directions, which can be approximated by the finite difference method.

[0064] Step 45: Construct the three components of the structure tensor based on the local gradient.

[0065] Construct the structure tensor components based on the local gradient: ; ; ; in, , and These represent the three components of the local structure tensor; The Gaussian smoothing window is the same as the smoothing window used in the aforementioned activity calculation. , and These are the outer integrals of the gradient vector. The structure tensor can simultaneously describe the gradient energy and orientation distribution characteristics of the local structure. Reflects the local total gradient energy. It reflects the correlation of gradient direction.

[0066] Step 46: Calculate structural orientation consistency based on the structural tensor. .

[0067] To measure whether a local structure has a definite directionality, structural directional consistency is calculated: ; in, This indicates consistency in structural orientation; To prevent extremely small positive numbers with a denominator of zero, the typical value is [value missing]. When a local region exhibits a distinct linear structure or a crack response with a consistent orientation, Larger (close to 1); when a local area has noise, speckle, or irregularly oriented texture, Smaller (close to 0). Directional consistency. The physical significance of this is that it measures the degree of consistency of local gradient directions. Linear structures (such as cracks) have significantly greater gradient energy in one direction than in other directions, thus exhibiting high directional consistency.

[0068] Step 47, calculate local structural energy .

[0069] Further calculations of local structural energy: ; in, This indicates a normalization process that maps the structure energy to... interval; This represents the local structural energy. Structural energy is used to measure whether there is a sufficiently strong edge or crack detail response at that location. Normalization methods can be min-max normalization or quantile normalization; in this embodiment, quantile normalization is preferred to enhance robustness to outliers.

[0070] Step 48: Combine structural energy and structural orientation consistency to obtain structural confidence level. .

[0071] Combining structural energy and structural orientation consistency, we obtain the structural confidence level: ; in, Indicates structural confidence; and These are weighting coefficients, which control the contribution of structural energy and directional consistency to the confidence score, respectively. In one specific implementation, they can be taken as... , ,Right now: ; in, The value can range from 0.20 to 0.50. The value can range from 0.50 to 0.80, preferably... , This formula shows that the structural orientation is consistent. The contribution weight to the confidence level (0.70) is greater than that to the structural energy. The baseline contribution (0.30) is because in ground-penetrating radar crack imaging, directional continuity is a better indicator of the true nature of the crack than simple amplitude intensity.

[0072] As can be seen from the above formula, structural confidence is constrained by both local structural energy and directional consistency. For noise responses with only strong amplitude but chaotic orientation, their structural directional consistency is low, thus suppressing structural confidence. For crack responses with moderate amplitude but continuous orientation and obvious linear characteristics, their structural confidence is high, and they can obtain higher weights in subsequent fusion.

[0073] For detail layers in the X direction respectively and Y-direction detail layer After executing steps 44 to 48, we obtain: ; ; in, and These represent the structural confidence levels of the detail layers in the X and Y directions, respectively. , and , The structural energy and directional consistency of the detail layers in the X and Y directions are respectively.

[0074] Step 5: Adaptive Determination of Detail Layer Blending Weights Activity map of detail layer in the X direction With the corresponding structural confidence level Joint construction X-direction detailed reliability score The activity map of the detail layer in the Y direction With the corresponding structural confidence level Joint construction Y-direction detailed reliability score Reliability scores based on details in the X and Y directions. , Adaptively determine the fusion weights of the X-direction detail layer and the Y-direction detail layer. , This allows for higher weighting of structurally continuous and directional crack details, resulting in a fused detail layer. .

[0075] Step 51: Combine the activity map with the structural confidence score to construct a detailed reliability score.

[0076] Combine CSR activity with structural confidence to construct a detailed reliability score: ; ; in, and These represent the reliability scores of the detail layers in the X and Y directions, respectively. and These are the normalized values ​​of the activity maps in the X and Y directions, respectively. and These are weighting adjustment coefficients, which respectively control the baseline contribution of activity and the contribution of structural confidence enhancement. In one specific implementation, a weighting coefficient of 1 can be used. , ,Right now: ; ; in, The value range can be 0.10 to 0.40. The value can range from 0.60 to 0.90, preferably... , .

[0077] Therefore, this invention does not determine the fusion weight solely based on local detail amplitude, but simultaneously considers detail response intensity and structural reliability. The reliability score consists of two parts: the activity normalized value. As a foundational term, it ensures that detailed responses of a certain strength can obtain basic weights; structural confidence. As an enhancement, only regions with high structural reliability receive additional weight increases. Only regions with both high activity and high structural confidence receive the highest weight during the fusion process.

[0078] Step 52: Construct the fusion weights for the X-direction detail layer and the Y-direction detail layer based on the detail reliability score.

[0079] Construct detail layer fusion weights in the X and Y directions based on the detail reliability score: ; ; in, and These represent the fusion weights of the detail layers in the X and Y directions, respectively, satisfying... ; This represents the weight enhancement factor, used to adjust the degree of influence of reliability score differences on the fusion weights; To prevent extremely small positive numbers with a denominator of zero, the typical value is [value missing]. .when When it is large (e.g.) The side with a higher reliability score will receive a more significant weighting advantage; when When smaller (e.g.) The fusion weight changes are relatively smooth. This embodiment preferably exhibits... The value can range from 1 to 4, ensuring the smoothness of the fusion result while giving appropriate weight to the high reliability region.

[0080] Step 53 yields the fusion detail layer under structural confidence constraints.

[0081] Finally, the fusion detail layer under structural confidence constraints is obtained: ; in, This represents the fused detail layer. This detail layer can synthesize the high-frequency response of the cracks in both the X and Y directions, and prioritizes the preservation of crack details with structural continuity, clear orientation, and high reliability.

[0082] Step 6: Information Entropy Weight Fusion of the Base Layer For the base layer in the X direction and Y-direction base layer Information entropy weight fusion is performed to obtain the fusion base layer. .

[0083] The base layer primarily reflects the low-frequency background, overall energy distribution, and large-scale variations in the C-scan image. To avoid disrupting the overall imaging structure during detail enhancement, this invention performs information entropy weighted fusion on the X-direction and Y-direction base layers.

[0084] Step 61, calculate the base layer in the X direction respectively. and Y-direction base layer Information entropy and .

[0085] Calculate the information entropy of the base layer separately: ; in, Represents the gray-level distribution of the base layer image. The probability of each gray level can be obtained through gray-level histogram statistics. This represents the information entropy of the base layer, measured in bits. The higher the information entropy, the richer the overall grayscale distribution information contained in that base layer, meaning the image has more grayscale levels and a greater amount of information.

[0086] Step 62: Determine the fusion weights of the base layer based on the information entropy.

[0087] The fusion weights of the base layer are determined based on the information entropy of the base layer in the X and Y directions: ; ; in, and These represent the base layer fusion weights in the X and Y directions, respectively, satisfying... ; To prevent extremely small positive numbers with a denominator of zero, the typical value is [value missing]. The physical significance of information entropy weight fusion lies in the fact that the base layer with higher information entropy contains richer grayscale distribution information and should be given a higher fusion weight, so as to retain more effective background structure information in the fusion result.

[0088] Step 63: The X-direction base layer and the Y-direction base layer are weighted and summed according to the base layer fusion weight to obtain the fused base layer.

[0089] The base layer fusion result is as follows: ; in, This represents the fusion base layer. By fusing information entropy at the base layer, the overall energy distribution and low-frequency background information can be preserved while avoiding the excessive dominance of unidirectional data in the fusion result.

[0090] Step 7: Image Reconstruction Will merge the base layer With the fusion detail layer Image reconstruction was performed to obtain the final crack-enhanced C-scan imaging results. .

[0091] The fusion base layer obtained in step 6 Blend detail layer obtained in step 5 Add them together to get the final merged image: ; in, This represents the final crack-enhanced C-scan image; This represents the detail gain coefficient, which is used to adjust the degree of enhancement of the crack detail layer in the final image. The value can be adjusted according to the actual imaging effect: when At that time, the details are enhanced, and the edges of the cracks are more prominent; when At this time, detail components are suppressed, resulting in a smoother image. In this embodiment, it is recommended... The range of values ​​is Preferred A balance is struck between enhancing crack details and maintaining image naturalness.

[0092] To demonstrate the effectiveness of the CSR fusion method based on structural confidence constraints of this invention, different fusion methods are compared with this method. The processed C-scan images are shown below. Figure 2 As shown.

[0093] Figure 2 In the middle (a), the C-scan image after average fusion is shown. Average fusion directly averages the data in the X and Y directions at the pixel level, which can retain some response information in both directions at the same time, but the fusion rules are relatively simple. It can be seen that the overall response of the crack or structure is relatively smooth, some boundaries are weakened, local details are not prominent enough, and the background texture is still obvious, indicating that average fusion is prone to causing the target information and background noise to be superimposed simultaneously. Figure 2 (b) shows the C-scan image after wavelet transform fusion. Wavelet transform fusion processes low-frequency and high-frequency information separately through multi-scale decomposition, which can enhance some edge and detail structures. The crack texture in the image is richer than that of average fusion, but the detail response is more fragmented, and there are more grid-like artifacts and high-frequency noise in some areas, resulting in insufficient target continuity. This shows that although wavelet fusion can improve the detail representation, it has weak constraints on background stripes and directional noise in GPR orthogonal data. Figure 2 Image (c) represents the C-scan image after local energy weighted fusion. Local energy weighted fusion assigns weights in the X / Y directions based on the local energy magnitude, thus highlighting areas with strong reflections. The main cracks and linear structures in the image are brighter, and the contrast of local targets is significantly improved. However, this method tends to favor high-energy responses, and strong background reflections and local high-energy artifacts are also simultaneously enhanced, resulting in excessively bright spots or bands in some areas, making it difficult to distinguish non-target structures from cracked targets. Figure 2 Image (d) represents the C-scan image fused by CSR based on structural confidence constraints according to this invention. This method introduces structural confidence constraints on top of CSR activity, comprehensively considering the intensity of local sparse details, directional consistency, and structural continuity. As shown in the image, the fusion result retains complementary information in the X / Y orthogonal directions, while exhibiting clearer fracture boundaries, better structural connectivity, a relatively smoother background region, and some suppression of isolated strong noise and discontinuous artifacts. Compared to the previous three methods, this method avoids the detail weakening of average fusion and reduces the fragmentation noise of wavelet fusion and the high-energy artifact enhancement problem of local energy-weighted fusion, making it more suitable for fracture-enhanced C-scan imaging.

[0094] To further verify the practical effect of the present invention, three existing methods—average fusion, wavelet transform fusion, and local energy weighted fusion—were compared with the present invention's method (CSR fusion based on structural confidence) on the same set of orthogonal GPR measured data. Four objective evaluation indicators—information entropy (IE), target-background contrast ratio (CNR), local signal-to-noise ratio (SNR), and background standard deviation (BG_STD)—were selected for quantitative evaluation. The results are shown in Table 1.

[0095] Table 1 Evaluation Indicators

[0096] As can be seen from Table 1, in terms of information entropy, the IE value of local energy weighted fusion is 7.2869, which is the highest among the four methods, indicating that its image grayscale distribution is the richest. However, the high index may also be due to the enhancement of noise texture, local high-energy artifacts or background fluctuations, and does not necessarily mean that the fusion effect is better. The IE value of the method of this invention is 6.2944, which is slightly lower than that of local energy weighted fusion, but it performs best in the other three key indicators: CNR reaches 12.9366, which is significantly higher than average fusion (7.7140), wavelet transform fusion (8.0840) and local energy weighted fusion (6.8951), indicating that the present invention can maximize the contrast between the fracture target and the background; SNR is 17.8637 dB, which is better than the other three methods (13.9572 dB, 14.3394 dB, and 12.7240 dB, respectively), reflecting the significant advantage of the present invention in suppressing noise; BG_STD is 8.2044, which is the lowest among all methods (the others are 12.5884, 10.9147, and 18.2877 respectively), proving that the present invention can effectively suppress background clutter and make the background area cleaner and more uniform.

[0097] Based on the above indicators, this invention outperforms average fusion, wavelet transform fusion, and local energy weighted fusion in terms of target contrast, signal-to-noise ratio, and background stability. It can effectively improve the imaging quality of orthogonal GPR crack-enhanced C-scan images, verifying the rationality and superiority of the structure confidence-constrained fusion strategy proposed in this invention.

[0098] II. System Implementation Examples This invention also provides an orthogonal GPR crack enhancement imaging system based on structural confidence constraints. This system corresponds one-to-one with the above-described method embodiments and can be implemented using software, hardware, or a combination of both. The system embodiments are described in detail below with reference to the accompanying drawings.

[0099] like Figure 3As shown, the system includes: a data acquisition module, a spatial registration module, a two-layer decomposition module, a structural confidence calculation module, a detail layer fusion module, a base layer fusion module, and an image reconstruction module.

[0100] The data acquisition module is used to acquire X-direction ground-penetrating radar data and Y-direction ground-penetrating radar data within the same detection area, and generate corresponding X-direction C-scan images and Y-direction C-scan images respectively. The X-direction and Y-direction are orthogonal to each other. The spatial registration module is used to unify the spatial coordinates and resample the physical grid of the C-scan image in the Y direction with the X-direction C-scan image as a reference, so that the two sets of C-scan images are in the same spatial coordinate system and under the same physical grid. The two-layer decomposition module is used to perform two-layer decomposition on the X-direction C-scan image and the Y-direction C-scan image after the spatial coordinates are unified, so as to obtain their respective base layer and detail layer; The structure confidence calculation module is used to calculate the activity maps based on convolutional sparse representation for the X-direction detail layer and the Y-direction detail layer respectively, and to calculate the structure confidence based on the local structure tensor respectively. The detail layer fusion module is used to jointly construct a detail reliability score by combining the activity map with the corresponding structural confidence. The fusion weight is adaptively determined based on the detail reliability scores in the X and Y directions to obtain the fused detail layer. The base layer fusion module is used to perform information entropy weight fusion on the X-direction base layer and the Y-direction base layer to obtain a fused base layer. The image reconstruction module is used to reconstruct the image by fusing the base layer and the detail layer to obtain the final crack-enhanced C-scan imaging result.

[0101] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily 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 invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An orthogonal GPR crack enhancement imaging method based on structural confidence constraints, characterized in that, Includes the following steps: Step 1: Acquire ground-penetrating radar data in the X and Y directions within the same detection area, and generate corresponding X and Y direction C-scan images respectively, wherein the X and Y directions are orthogonal to each other; Step 2: Using the C-scan image in the X direction as a reference, perform spatial coordinate unification and physical grid resampling on the C-scan image in the Y direction, so that the two sets of C-scan images are in the same spatial coordinate system and under the same physical grid. Step 3: Perform two-layer decomposition on the C-scan images in the X and Y directions after unifying the spatial coordinates to obtain their respective base layer and detail layer; Step 4: Perform directional filtering and sparsification based on convolutional sparse representation on the detail layers in the X and Y directions respectively to obtain the activity maps of the detail layers in the X and Y directions; at the same time, calculate the structure confidence of the detail layers in the X and Y directions respectively. Step 5: Construct detail reliability scores for the X and Y directions by combining the activity maps of the detail layers in the X and Y directions with the corresponding structural confidence scores. Adaptively determine the fusion weights of the detail layers in the X and Y directions based on the detail reliability scores in the X and Y directions to obtain the fused detail layers. Step 6: Perform information entropy weight fusion on the base layers in the X and Y directions to obtain the fused base layer; Step 7: Perform image reconstruction on the fused base layer and the fused detail layer to obtain the final crack-enhanced C-scan imaging result.

2. The method according to claim 1, characterized in that, Step 1, which generates C-scan images in the X and Y directions, specifically includes: Step 11, let the X-direction ground-penetrating radar data be... Y-direction ground-penetrating radar data is ,in, Indicates the time sampling point. and These represent two horizontal spatial directions in the same physical coordinate system. The current C-scan imaging time window is denoted as... ; Step 12: Perform local Gaussian averaging on the original ground-penetrating radar data to remove the background, and obtain the background-removed radar data. Step 13: Perform Kirchhoff migration imaging processing on the background-removed ground-penetrating radar data; Step 14, after Kirchhoff offset processing, within the selected time window Extract C-scan images in the X direction separately. and C-scan image in the Y direction .

3. The method according to claim 1, characterized in that, In step 2, the spatial coordinates of the C-scan image in the Y direction are unified and the physical grid is resampled, specifically including: Step 21: Transpose the C-scan image in the Y direction so that its matrix axis is consistent with the C-scan image in the X direction; Step 22: Perform flipping and translation correction on the transposed Y-direction C-scan image to obtain a spatially aligned Y-direction image; Step 23: Establish a unified physical grid based on the actual sampling interval of the X and Y direction data, and resample the two sets of C-scan images to the same spatial scale to obtain the resampled X-direction C-scan image. and C-scan image in the Y direction .

4. The method according to claim 1, characterized in that, In step 3, the two-layer decomposition uses the Tikhonov decomposition method. Taking the X-direction image as an example, the base layer is obtained through frequency domain low-pass filtering, and the detail layer is obtained by subtracting the base layer from the original image, i.e.: ； ； in, and These represent the base layers of the images in the X and Y directions, respectively, and contain low-frequency background, overall energy distribution, and gradually varying regions. and These represent detail layers of the images in the X and Y directions, respectively, and include crack edges, local mutations, high-frequency textures, and directional structural responses.

5. The method according to claim 1, characterized in that, In step 4, directional filtering and sparsification based on convolutional sparse representation are performed on the detail layers in the X and Y directions to obtain the activity maps of the detail layers in the X and Y directions, specifically including: Step 41: Perform convolution filtering on the X-direction detail layer and the Y-direction detail layer respectively using a directional filter bank to obtain the corresponding filter response; Step 42: Perform soft threshold sparsification on the filter response in each direction to obtain a sparse coefficient map; Step 43: Calculate the activity map of the detail layer in the X direction based on the sparse coefficient map. Activity plot of detail layer in the Y direction .

6. The method according to claim 1, characterized in that, Step 4, calculating the structural confidence of the detail layers in the X and Y directions, specifically includes: Step 44, for the detail layer in any direction Calculate its horizontal and vertical gradients: ； Step 45: Construct the three components of the structure tensor based on the local gradient: ； ； ； in, Use a Gaussian smoothed window. This represents the convolution operation. , , These are the three components of the local structure tensor, used to describe the gradient energy and orientation distribution characteristics of the local structure; Step 46: Calculate structural orientation consistency based on the structural tensor: ； in, To ensure structural consistency, To prevent extremely small positive numbers with a denominator of zero; Step 47, calculate the local structural energy: ； in, For local structural energy, This indicates normalization processing; Step 48: Combine structural energy and structural orientation consistency to obtain structural confidence level: ； in, For structural confidence, and These are the weighting coefficients; For detail layers in the X direction respectively and Y-direction detail layer Perform steps 44 to 48 to obtain the structural confidence of the detail layer in the X direction. and the structural confidence of the detail layer in the Y direction .

7. The method according to claim 1, characterized in that, Step 5 specifically includes: Step 51: Combine the activity map with the structural confidence score to construct a detailed reliability score. ； ； in, , The reliability scores for the detail layers in the X and Y directions are respectively. , These are the activity maps for the detail layers in the X and Y directions, respectively. , These represent the structural confidence levels of the detail layers in the X and Y directions, respectively. and This is the weighting adjustment coefficient. This indicates normalization processing; Step 52: Construct the fusion weights for the X-direction detail layer and the Y-direction detail layer based on the detail reliability score. ； ； in, , These are the fusion weights for the detail layers in the X and Y directions, respectively. As a weighting enhancement factor, To prevent extremely small positive numbers with a denominator of zero; Step 53, obtaining the fusion detail layer under structural confidence constraints: ； in, To integrate the detail layer, , These are detail layers for the X and Y directions, respectively.

8. The method according to claim 1, characterized in that, Step 6 specifically includes: Step 61, calculate the base layer in the X direction respectively. and Y-direction base layer Information entropy and ; Step 62: Determine the base layer fusion weights based on information entropy: ； ； in, , These are the fusion weights of the base layers in the X and Y directions, respectively. , These are the information entropies of the base layer in the X and Y directions, respectively. To prevent extremely small positive numbers with a denominator of zero; Step 63: Weight the X-direction base layer and the Y-direction base layer according to the base layer fusion weights to obtain the fused base layer: ； in, To integrate the base layer.

9. The method according to claim 1, characterized in that, In step 7, the fusion base layer and the fusion detail layer are added together to obtain the final crack-enhanced C-scan imaging result: ； in, For the final crack-enhanced C-scan image, To integrate the base layer, To integrate the detail layer, This is the detail gain coefficient, used to adjust the enhancement level of the crack detail layer in the final image.

10. An orthogonal GPR crack enhancement imaging system based on structural confidence constraints, characterized in that, include: The data acquisition module is used to acquire X-direction ground-penetrating radar data and Y-direction ground-penetrating radar data within the same detection area, and generate corresponding X-direction C-scan images and Y-direction C-scan images respectively, wherein the X-direction and the Y-direction are orthogonal to each other; The spatial registration module is used to unify the spatial coordinates and resample the physical grid of the C-scan image in the Y direction with the X-direction C-scan image as a reference, so that the two sets of C-scan images are in the same spatial coordinate system and under the same physical grid. The two-layer decomposition module is used to perform two-layer decomposition on the C-scan image in the X direction and the C-scan image in the Y direction after the spatial coordinates are unified, so as to obtain their respective base layer and detail layer; The structure confidence calculation module is used to calculate the activity maps based on convolutional sparse representation for the X-direction detail layer and the Y-direction detail layer respectively, and to calculate the structure confidence based on the local structure tensor respectively. The detail layer fusion module is used to jointly construct a detail reliability score by combining the activity map with the corresponding structural confidence score, and adaptively determine the fusion weight based on the detail reliability scores in the X and Y directions to obtain the fused detail layer. The base layer fusion module is used to perform information entropy weight fusion on the X-direction base layer and the Y-direction base layer to obtain a fused base layer. The image reconstruction module is used to reconstruct the image of the fused base layer and the fused detail layer to obtain the final crack-enhanced C-scan imaging result.

Images