Ground penetrating radar image cavity imaging method suitable for different geological conditions

Abstract

The invention discloses a ground penetrating radar image cavity imaging method suitable for different geological conditions, and belongs to the technical field of ground penetrating radar detection. In order to solve the problem that a ground penetrating radar signal processing technology cannot effectively adapt to dynamic changes of medium parameters under different geological conditions, the method comprises the following steps: constructing a normalized field intensity which dynamically changes along with depth and is subjected to background suppression; phase consistency structural strength is extracted; obtaining a depth-medium coupling potential energy field; constructing an anisotropic Gaussian kernel function with direction selectivity; calculating a thermodynamic upper bound manifold projection and a thermodynamic lower bound manifold projection; extracting an enhanced gradient feature field by calculating the difference between the thermodynamic upper bound manifold projection and the thermodynamic lower bound manifold projection; and establishing multi-scale entropy weight fusion inverse time imaging to obtain a final underground cavity imaging result. According to the invention, detection images with high signal-to-noise ratios and clear outlines can be obtained under different geological conditions.

Description

Technical Field

[0001] This invention belongs to the field of ground-penetrating radar detection technology, specifically relating to a ground-penetrating radar image cavity imaging method adapted to different geological conditions. Background Technology

[0002] While ground-penetrating radar (GPR) is the mainstream non-destructive testing method for detecting underground concealed works, its imaging quality is highly susceptible to the complexity and variability of different geological conditions within the detection area. In practical engineering scenarios, due to the uneven distribution of water content in underground media, differences in the gradation of roadbed materials, and the interlacing of strata, the echo signals received by radar often exhibit nonlinear attenuation characteristics and strong background clutter interference. Existing conventional data processing methods tend to employ globally uniform amplitude threshold truncation or linear frequency domain filtering algorithms. This lack of adaptive processing logic is ineffective when facing diverse geological environments: it easily misjudges background noise as anomalies in shallow, high-reflection areas and is prone to missing cavities in deep, high-attenuation areas due to weak signals. Therefore, traditional methods relying solely on amplitude strength or phase continuity are insufficient to meet the engineering requirements for accurate location and morphological reconstruction of underground cavities under varying geological conditions. Summary of the Invention

[0003] The present invention addresses the problem that ground-penetrating radar signal processing technology cannot effectively adapt to the dynamic changes in medium parameters under different geological conditions, and proposes a ground-penetrating radar image cavity imaging method that adapts to different geological conditions.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for imaging cavities in ground-penetrating radar images that is adaptable to different geological conditions includes the following steps:

[0006] S1. Ground penetrating radar collects raw underground data, and based on statistical principles, constructs a normalized field strength after background suppression that dynamically changes with depth;

[0007] S2. Considering the exponential amplitude attenuation of electromagnetic waves as they propagate underground, phase consistency analysis technology is introduced to extract the phase consistency structural strength.

[0008] S3. Couple the normalized field strength after background suppression obtained in step S1 with the phase-consistent structural strength obtained in step S2 to obtain the depth-medium coupled potential energy field.

[0009] S4. Construct an anisotropic Gaussian kernel function with direction selectivity, wherein the anisotropic Gaussian kernel function with direction selectivity is mathematically represented as an elliptical Gaussian distribution with a specific aspect ratio and rotation angle.

[0010] S5. Based on the deep-medium coupled potential energy field obtained in step S3 and the anisotropic Gaussian kernel function with direction selectivity obtained in step S4, calculate the thermodynamic upper boundary manifold projection and the thermodynamic lower boundary manifold projection.

[0011] S6. Extract the enhanced gradient feature field by calculating the difference between the thermodynamic upper boundary manifold projection and the thermodynamic lower boundary manifold projection;

[0012] S7. Based on the information entropy-based adaptive weighting strategy, combined with the enhanced gradient feature field obtained in step S6, a multi-scale entropy weight fusion reverse time imaging is established to obtain the final underground cavity imaging result.

[0013] Furthermore, in step S1, a background model reflecting the electromagnetic characteristics of the current geological strata is established. By combining a nonlinear mapping function, the anomalous response in the original signal and background fluctuations are effectively decoupled to obtain the normalized field strength after background suppression. .

[0014] Furthermore, step S2 involves constructing a multi-scale filter bank to perform frequency domain decomposition and reconstruction of the radar profile, obtaining the phase-consistent structural strength, expressed as:

[0015]

[0016] in, For phase-consistent structural strength, x is the horizontal survey line position index, z is the vertical depth position index, obtained from the raw data collected by ground penetrating radar; n is the filter scale index, corresponding to cavity targets of different sizes, set by technicians according to references; The total number of dimensions is determined by professionals through preliminary experiments based on the target cavity size range and radar bandwidth. The frequency spread weights for the nth scale are preset by professionals based on the bandwidth or energy distribution of the filters at each scale. The signal envelope amplitude at the nth scale is obtained by performing a Hilbert transform on the filtered signal. The phase deviation weighting factor at the nth scale is determined by the instantaneous phase at the nth scale. with scale-averaged phase The deviation between them was calculated. ; The noise estimation threshold is determined by expert experience.

[0017] Furthermore, a depth compensation mechanism is introduced in step S3, resulting in:

[0018]

[0019] in, This represents the potential energy field value of deep-medium coupling. The weighting coefficients are determined by expert experience. This is a media attenuation compensation factor, set by experts based on geological type.

[0020] Furthermore, the expression for the direction-selective anisotropic Gaussian kernel function constructed in step S4 is as follows:

[0021]

[0022] in, For the scale s direction The kernel weight; s is the scale index, corresponding to hollow targets of different sizes, which is determined by professionals; The rotation angle of the kernel function is determined by professionals, representing the horizontal, oblique, or vertical direction. These are the row and column indices in the local coordinate system of the kernel function, defined within the finite support domain of the convolution kernel, and determined by professionals. , These are the standard deviations of the anisotropic Gaussian kernel along the principal and secondary axes, respectively, which determine the kernel's aspect ratio and spatial resolution. These values ​​are set by professionals based on s.

[0023] Furthermore, step S5 combines convolution operations and exponential mapping to calculate the thermodynamic upper bound manifold projection, using the following formula:

[0024]

[0025] in, The upper bound is the scalar value of the manifold; The temperature reciprocal coefficient controls the sensitivity to local extrema and is determined by expert experience; D is the integration domain, which is the spatial range covered by the kernel function and is determined by professionals based on the scale index s.

[0026] The projection of the thermodynamic lower boundary manifold is calculated by employing a negative exponential mapping transformation. The calculation formula is as follows:

[0027]

[0028] in, It is the lower bound scalar value of the manifold.

[0029] Furthermore, the expression for the enhanced gradient feature field extracted in step S6 is as follows:

[0030]

[0031] in, The enhanced gradient feature field; , These are the second-order partial derivatives along the x and y directions, respectively; , These are the first-order partial derivatives along the x and y directions, respectively.

[0032] Furthermore, in step S7, the effective information content of the image is evaluated by calculating the information entropy of the image at each scale. The final underground cavity imaging result is then output by weighted summation and projection onto the maximum value of the direction, expressed as:

[0033]

[0034] in, This is the final imaging result of the underground cavity; Gradient field at scale s Image information entropy, Gradient field at the k-th scale The image information entropy is determined by professionals. This represents the maximum value in the direction.

[0035] The beneficial effects of this invention are:

[0036] The present invention provides a ground-penetrating radar image cavity imaging method adapted to different geological conditions. This method overcomes the problems of missed detection of deep targets and false positives in shallow layers caused by insufficient extraction of single features and lack of flexibility in fixed threshold segmentation in traditional methods. Thus, it provides an adaptive detection scheme that can take into account both shallow interference suppression and deep weak signal enhancement.

[0037] The present invention discloses a ground-penetrating radar image cavity imaging method adapted to different geological conditions. First, a normalized field adapted to changes in geological background is established and phase consistency features sensitive to abrupt changes in the medium are extracted. Background energy and structural features are fused to construct a deeply coupled potential energy field, and directional detection is performed using anisotropic kernel functions. On this basis, thermodynamic and statistical mechanics concepts are introduced to calculate the manifold projection of the potential energy field, and edge gradients are enhanced through difference and curvature operations. Finally, the contribution of each component is weighed based on multi-scale information entropy, and the final image is generated by fusion.

[0038] This invention discloses a ground-penetrating radar (GPR) image cavity imaging method adapted to different geological conditions. By constructing a deep-coupled potential energy field and a manifold projection algorithm, it effectively solves the problems of nonlinear attenuation and noise interference of electromagnetic waves in complex media. This method utilizes a physical compensation mechanism to achieve balanced detection across the entire depth and combines a multi-scale information entropy fusion strategy to dynamically adjust weights. This significantly sharpens target edges while suppressing random geological noise, ensuring high signal-to-noise ratio and clear contour detection imaging under various geological conditions. Attached Figure Description

[0039] Figure 1 This is a flowchart of a ground-penetrating radar image cavity imaging method adapted to different geological conditions, as described in this invention.

[0040] Figure 2 For comparison, simulated ground-penetrating radar images of the interior of roads using existing technology;

[0041] Figure 3 This is the result image after processing by the ground-penetrating radar image cavity imaging method adapted to different geological conditions as described in this invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described specific embodiments are merely a part of the embodiments of the invention, and not all of them. The components of the specific embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations, and the invention may also have other embodiments.

[0043] Therefore, the following detailed description of specific embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected specific embodiments of the invention. All other specific embodiments obtained by those skilled in the art based on these specific embodiments without inventive effort are within the scope of protection of this invention.

[0044] To further understand the invention's content, features, and effects, the following specific embodiments are provided, along with accompanying drawings. Figure 1 -Appendix Figure 3 Detailed explanation is as follows:

[0045] Example 1:

[0046] A method for imaging cavities in ground-penetrating radar images that is adaptable to different geological conditions includes the following steps:

[0047] S1. Ground penetrating radar collects raw underground data, and based on statistical principles, constructs a normalized field strength after background suppression that dynamically changes with depth;

[0048] Furthermore, in step S1, a background model reflecting the electromagnetic characteristics of the current geological strata is established. By combining a nonlinear mapping function, the anomalous response in the original signal and background fluctuations are effectively decoupled to obtain the normalized field strength after background suppression. The expression is:

[0049]

[0050] in, is the normalized field strength after background suppression; x is the horizontal survey line position index, and z is the vertical depth position index, obtained from the raw data collected by ground penetrating radar; The original ground-penetrating radar echo amplitude is obtained from the raw data collected by the radar equipment. This is the background mean value at the current depth layer, obtained from the raw data collected by the radar equipment; This represents the background standard deviation at the current depth layer, obtained from the raw data collected by the radar equipment. The signal-to-noise ratio adjustment coefficient can be selected by adjusting multiple sets of experimental data in a background region without holes, so as to achieve the optimal effect of background noise suppression and target fidelity. The ambient noise floor threshold can be determined by statistical analysis in a background area without obvious reflectors. The probability distribution is used, and its mean or the mean plus a certain number of standard deviations is taken as the threshold.

[0051] Furthermore, in actual ground-penetrating radar (GPR) data acquisition, the significant differences in dielectric constants of shallow subsurface media (such as asphalt pavement, backfill, and undisturbed soil) lead to a highly non-uniform distribution of the background baseline of the radar echo signal. To eliminate the interference of this geological heterogeneity on subsequent imaging, this step constructs a dynamically changing, depth-adaptive normalized geological background field based on statistical principles. This method first calculates the statistical moments within a sliding window along the depth direction to establish a background model that reflects the electromagnetic characteristics of the current geological stratum. Then, through a nonlinear mapping function, it effectively decouples the anomalous response in the original signal from background fluctuations. This processing method ensures that the signal baseline can be aligned to a uniform dimensional range, regardless of whether it is in a high-water-content clay layer or a dry sand and gravel layer.

[0052] S2. Considering the exponential amplitude attenuation of electromagnetic waves as they propagate underground, phase consistency analysis technology is introduced to extract the phase consistency structural strength.

[0053] Furthermore, step S2 involves constructing a multi-scale filter bank to perform frequency domain decomposition and reconstruction of the radar profile, obtaining the phase-consistent structural strength, expressed as:

[0054]

[0055] in, For phase-consistent structural strength, x is the horizontal survey line position index, z is the vertical depth position index, obtained from the raw data collected by ground penetrating radar; n is the filter scale index, corresponding to cavity targets of different sizes, set by technicians according to references; The total number of dimensions is determined by professionals through preliminary experiments based on the target cavity size range and radar bandwidth. The frequency spread weights for the nth scale are preset by professionals based on the bandwidth or energy distribution of the filters at each scale. The signal envelope amplitude at the nth scale is obtained by performing a Hilbert transform on the filtered signal. The phase deviation weighting factor at the nth scale is determined by the instantaneous phase at the nth scale. with scale-averaged phase The deviation between them was calculated. ; The noise estimation threshold is determined by expert experience.

[0056] Furthermore, the instantaneous phase at the nth scale with scale-averaged phase The calculation formula is:

[0057]

[0058]

[0059] in, It is the arctangent function; The imaginary part of the nth-scale filtering result is obtained by convolving the original acquired radar data with the imaginary part of the nth-scale filter kernel. These are the real parts of the filtering result at the nth scale, which are obtained by convolving the original acquired radar data with the real parts of the nth scale filter kernel.

[0060] Furthermore, given that electromagnetic waves experience exponential amplitude attenuation with increasing depth during underground propagation, detection methods relying solely on amplitude intensity often struggle to identify minute cavities deep within the earth. However, cavities, as a type of abrupt change in medium, exhibit phase discontinuities at their boundaries that are more robust to medium attenuation than amplitude characteristics. Therefore, this step introduces phase consistency analysis techniques, constructing a multi-scale filter bank to perform frequency domain decomposition and reconstruction of the radar profile. This process aims to capture the phase alignment of each frequency component in the signal at specific locations. This alignment accurately characterizes the geometric structure information of the cavity's top and bottom plates, unaffected by signal energy intensity. By calculating the phase structure characteristic tensor, even in deep, low signal-to-noise ratio regions, the target's skeletal outline can be clearly delineated, compensating for the limitations of energy fields in deep detection capabilities.

[0061] S3. Couple the normalized field strength after background suppression obtained in step S1 with the phase-consistent structural strength obtained in step S2 to obtain the depth-medium coupled potential energy field.

[0062] Furthermore, a depth compensation mechanism is introduced in step S3, resulting in:

[0063]

[0064] in, This represents the potential energy field value of deep-medium coupling. The weighting coefficients are determined by expert experience. This is a media attenuation compensation factor, set by experts based on geological type.

[0065] Furthermore, to achieve a comprehensive description of the underground cavity target, the background energy information obtained in step S1 must be fused with the geometric structure information obtained in step S2. This step defines a depth-medium coupled potential energy field, which not only linearly combines energy and structural features but, more importantly, introduces a depth compensation mechanism. Considering the different absorption rates of electromagnetic waves by different geological media, this step introduces a depth gain exponent based on the medium type to physically compensate for the deep signal. The initial design of this formula is to elevate the weak deep cavity signal to the same order of magnitude as the shallow signal, thereby constructing a unified potential energy space with clear physical meaning and insensitive to depth changes. Within this space, both shallow and deep cavities exhibit high potential energy anomalous regions, providing a unified input benchmark for subsequent manifold projection.

[0066] S4. Construct an anisotropic Gaussian kernel function with direction selectivity, wherein the anisotropic Gaussian kernel function with direction selectivity is mathematically represented as an elliptical Gaussian distribution with a specific aspect ratio and rotation angle.

[0067] Furthermore, the expression for the direction-selective anisotropic Gaussian kernel function constructed in step S4 is as follows:

[0068]

[0069] in, For the scale s direction The kernel weight; s is the scale index, corresponding to hollow targets of different sizes, which is determined by professionals; The rotation angle of the kernel function is determined by professionals, representing the horizontal, oblique, or vertical direction. These are the row and column indices in the local coordinate system of the kernel function, defined within the finite support domain of the convolution kernel, and determined by professionals. , These are the standard deviations of the anisotropic Gaussian kernel along the principal and secondary axes, respectively, which determine the kernel's aspect ratio and spatial resolution. These values ​​are set by professionals based on s.

[0070] Furthermore, the shapes of underground cavities vary greatly, ranging from regular circular pipes to irregular collapsed cavities, and often exhibit hyperbolic or specific directional textures on radar profiles. To accurately capture these morphological features from the potential energy field, this step constructs a set of direction-selective anisotropic Gaussian kernel functions. Mathematically, these kernel functions are elliptical Gaussian distributions with specific aspect ratios and rotation angles. They act like probes with specific orientations, enabling multi-angle scanning and convolution of the potential energy field in subsequent steps. By adjusting the rotation angle and scale parameters, this kernel function can adapt to different cavity targets, thus preserving the target edges while effectively suppressing random noise using the smoothing properties of Gaussian functions, providing a directional convolution operator for manifold projection calculations.

[0071] S5. Based on the deep-medium coupled potential energy field obtained in step S3 and the anisotropic Gaussian kernel function with direction selectivity obtained in step S4, calculate the thermodynamic upper boundary manifold projection and the thermodynamic lower boundary manifold projection.

[0072] Furthermore, step S5 combines convolution operations and exponential mapping to calculate the thermodynamic upper bound manifold projection, using the following formula:

[0073]

[0074] in, The upper bound is the scalar value of the manifold; The temperature reciprocal coefficient controls the sensitivity to local extrema and is determined by expert experience; D is the integration domain, which is the spatial range covered by the kernel function and is determined by professionals based on the scale index s.

[0075] To extract salient features representing the presence of voids from complex potential energy fields, this step introduces the concept of a partition function from statistical thermodynamics to calculate the upper bound manifold projection of the potential energy field. Mathematically, this process can be understood as a soft max-pooling operation, which convolves the probe kernel generated in step four with the potential energy field generated in step three, and amplifies the contribution of high-energy regions through exponential mapping. The upper bound manifold primarily reflects the strongest response envelope of the signal within a local region, effectively connecting discrete strong reflection points into continuous energy clusters, thus outlining the main range of the potential void target. Unlike traditional linear smoothing, this exponentially weighted projection method can greatly preserve the peak characteristics of strong signals, preventing the target signal from being excessively blurred during the smoothing process.

[0076] The projection of the thermodynamic lower boundary manifold is calculated by employing a negative exponential mapping transformation. The calculation formula is as follows:

[0077]

[0078] in, It is the lower bound scalar value of the manifold.

[0079] Furthermore, corresponding to the upper bound manifold, the lower bound manifold projection aims to capture the background floor or low-energy envelope in the potential energy field. By employing a negative exponential mapping transformation, this formula focuses on weak signal characteristics within local regions. The purpose of calculating the lower bound manifold is to establish a local reference surface for subsequent differential comparison with the upper bound manifold. This dual-manifold design automatically cancels out large-scale, gradually varying background trends, thereby ensuring that the final detected target is an anomaly with a significant abrupt change relative to the local background.

[0080] S6. Extract the enhanced gradient feature field by calculating the difference between the thermodynamic upper boundary manifold projection and the thermodynamic lower boundary manifold projection;

[0081] After obtaining the upper and lower bound manifolds, this step extracts the saliency of the target by calculating their difference. However, relying solely on the difference may retain some high-frequency noise. Therefore, this step employs a curvature enhancement factor. The boundary of a cavity typically exhibits greater curvature on the manifold surface, while flat noise regions have smaller curvature. By calculating the second derivative of the upper bound manifold, a curvature weight term is constructed to nonlinearly modulate the manifold difference result. This operation significantly enhances the contrast of the cavity edge while suppressing flat non-target regions, achieving high signal-to-noise ratio feature extraction. This curvature-based enhancement mechanism essentially utilizes the sharpness of the cavity target's geometric shape to effectively distinguish it from gentle geological background noise.

[0082] Furthermore, the expression for the enhanced gradient feature field extracted in step S6 is as follows:

[0083]

[0084] in, The enhanced gradient feature field; , These are the second-order partial derivatives along the x and y directions, respectively; , These are the first-order partial derivatives along the x and y directions, respectively.

[0085] S7. Based on the information entropy-based adaptive weighting strategy, combined with the enhanced gradient feature field obtained in step S6, a multi-scale entropy weight fusion reverse time imaging is established to obtain the final underground cavity imaging result.

[0086] In the final stage, to obtain a clear, intuitive, and non-redundant map of underground cavities, gradient fields at different scales (corresponding to cavities of different sizes) and in different directions (corresponding to different shapes) need to be fused. This step employs an adaptive weighting strategy based on information entropy. By calculating the information entropy of the image at each scale, the effective information it contains is evaluated: the lower the entropy value, the clearer the image and the less noise it contains, thus assigning it a higher weight. Finally, through weighted summation and projection onto the maximum value of the direction, a two-dimensional matrix reflecting the probability of the existence of underground cavities is output, completing the final transformation from data to image. This step not only achieves effective complementarity of multi-source information but also automatically eliminates low-quality scale components through the entropy weighting mechanism, ensuring that the final imaging result has the highest confidence level.

[0087] Furthermore, in step S7, the effective information content of the image is evaluated by calculating the information entropy of the image at each scale. The final underground cavity imaging result is then output by weighted summation and projection onto the maximum value of the direction, expressed as:

[0088]

[0089] in, This is the final imaging result of the underground cavity; Gradient field at scale s Image information entropy, Gradient field at the k-th scale The image information entropy is determined by professionals. The maximum value in the direction;

[0090]

[0091]

[0092] Where i represents the gray level, which is set by professionals; For the gradient field at the s-th scale In the diagram, the probability of the i-th gray level is obtained by statistically analyzing the gradient field. The corresponding grayscale histogram is automatically calculated by dividing it by the total number of pixels. For the gradient field at the k-th scale In the diagram, the probability of the i-th gray level is obtained by statistically analyzing the gradient field. The corresponding grayscale histogram is automatically calculated by dividing it by the total number of pixels.

[0093] This embodiment ultimately calculates a two-dimensional probability matrix that integrates multi-scale gradient features and directional information. The numerical points in this matrix quantify the confidence level of the existence of cavitation anomalies at the corresponding underground spatial locations. Based on this, engineers can directly determine the burial depth, range, and geometric outline of underground cavities, thus using high-probability areas as the basis for subsequent excavation or grouting reinforcement decisions. This achieves trenchless, accurate assessment without the need for large-scale borehole sampling.

[0094] The following is an example illustrating its application:

[0095] This method is used to process simulated ground-penetrating radar images of the road interior. The processing results are compared to, for example, 2 and... Figure 3 As shown, the image processed by this method can accurately capture the location of the cavity region, demonstrating the effectiveness of the proposed method.

[0096] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0097] Although this application has been described above with reference to specific embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of this application. In particular, as long as there is no structural conflict, the features in the specific embodiments disclosed in this application can be combined with each other in any way. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, this application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for imaging cavities in ground-penetrating radar images adapted to different geological conditions, characterized in that, Includes the following steps: S1. Ground penetrating radar collects raw underground data, and based on statistical principles, constructs a normalized field strength after background suppression that dynamically changes with depth; S2. Considering the exponential amplitude attenuation of electromagnetic waves as they propagate underground, phase consistency analysis technology is introduced to extract the phase consistency structural strength. S3. Couple the normalized field strength after background suppression obtained in step S1 with the phase-consistent structural strength obtained in step S2 to obtain the depth-medium coupled potential energy field. S4. Construct an anisotropic Gaussian kernel function with direction selectivity, wherein the anisotropic Gaussian kernel function with direction selectivity is mathematically represented as an elliptical Gaussian distribution with a specific aspect ratio and rotation angle. S5. Based on the deep-medium coupled potential energy field obtained in step S3 and the anisotropic Gaussian kernel function with direction selectivity obtained in step S4, calculate the thermodynamic upper boundary manifold projection and the thermodynamic lower boundary manifold projection. S6. Extract the enhanced gradient feature field by calculating the difference between the thermodynamic upper boundary manifold projection and the thermodynamic lower boundary manifold projection; S7. Based on the information entropy-based adaptive weighting strategy, combined with the enhanced gradient feature field obtained in step S6, a multi-scale entropy weight fusion reverse time imaging is established to obtain the final underground cavity imaging result.

2. The ground-penetrating radar image cavity imaging method adapted to different geological conditions according to claim 1, characterized in that, In step S1, a background model reflecting the electromagnetic characteristics of the current geological strata is established. By combining a nonlinear mapping function, the anomalous response in the original signal and background fluctuations are effectively decoupled to obtain the normalized field strength after background suppression. .

3. The ground-penetrating radar image cavity imaging method adapted to different geological conditions according to claim 2, characterized in that, Step S2 involves constructing a multi-scale filter bank to perform frequency domain decomposition and reconstruction of the radar profile, obtaining the phase-consistent structure strength, expressed as: ； in, For phase-consistent structural strength, x is the horizontal survey line position index, z is the vertical depth position index, obtained from the raw data collected by ground penetrating radar; n is the filter scale index, corresponding to cavity targets of different sizes, set by technicians according to references; The total number of dimensions is determined by professionals through preliminary experiments based on the target cavity size range and radar bandwidth. The frequency spread weights for the nth scale are preset by professionals based on the bandwidth or energy distribution of the filters at each scale. The signal envelope amplitude at the nth scale is obtained by performing a Hilbert transform on the filtered signal. The phase deviation weighting factor at the nth scale is determined by the instantaneous phase at the nth scale. with scale-averaged phase The deviation between them was calculated. ; The noise estimation threshold is determined by expert experience.

4. The ground-penetrating radar image cavity imaging method adapted to different geological conditions according to claim 3, characterized in that, In step S3, a depth compensation mechanism is introduced, resulting in: ； in, This represents the potential energy field value of deep-medium coupling. The weighting coefficients are determined by expert experience. This is a media attenuation compensation factor, set by experts based on geological type.

5. The ground-penetrating radar image cavity imaging method adapted to different geological conditions according to claim 4, characterized in that, The expression for the direction-selective anisotropic Gaussian kernel function constructed in step S4 is as follows: ； in, For the scale s direction The kernel weight; s is the scale index, corresponding to hollow targets of different sizes, which is determined by professionals; The rotation angle of the kernel function is determined by professionals, representing the horizontal, oblique, or vertical direction. These are the row and column indices in the local coordinate system of the kernel function, defined within the finite support domain of the convolution kernel, and determined by professionals. , These are the standard deviations of the anisotropic Gaussian kernel along the principal and secondary axes, respectively, which determine the kernel's aspect ratio and spatial resolution. These values ​​are set by professionals based on s.

6. The ground-penetrating radar image cavity imaging method adapted to different geological conditions according to claim 5, characterized in that, Step S5 combines convolution operations and exponential mapping to calculate the thermodynamic upper bound manifold projection. The calculation formula is as follows: ； in, The upper bound is the scalar value of the manifold; The temperature reciprocal coefficient controls the sensitivity to local extrema and is determined by expert experience; D is the integration domain, which is the spatial range covered by the kernel function and is determined by professionals based on the scale index s. The projection of the thermodynamic lower boundary manifold is calculated by employing a negative exponential mapping transformation. The calculation formula is as follows: ； in, It is the lower bound scalar value of the manifold.

7. The ground-penetrating radar image cavity imaging method adapted to different geological conditions according to claim 6, characterized in that, The expression for extracting the enhanced gradient feature field in step S6 is as follows: ； in, The enhanced gradient feature field; , These are the second-order partial derivatives along the x and y directions, respectively; , These are the first-order partial derivatives along the x and y directions, respectively.

8. The ground-penetrating radar image cavity imaging method adapted to different geological conditions according to claim 7, characterized in that, In step S7, the effective information content of the image is evaluated by calculating the information entropy of the image at each scale. The final underground cavity imaging result is then output by weighted summation and projection onto the maximum directional value. The expression is: ； in, This is the final imaging result of the underground cavity; Gradient field at scale s Image information entropy, Gradient field at the k-th scale The image information entropy is determined by professionals. This represents the maximum value in the direction.

Images