A UAV-GPR noise suppression and imaging method and device based on Hankel trajectory matrix decomposition

Abstract

This invention discloses a UAV-GPR noise suppression and imaging method and apparatus based on Hankel trajectory matrix decomposition, belonging to the field of ground-penetrating radar technology. The method first preprocesses the radar echo data and constructs a Hankel-block-Hankel trajectory matrix, achieving noise suppression through singular value decomposition and adaptive truncation and reconstruction based on a quasi-local curvature metric. Then, the derivative threshold reference point method is used to accurately extract the arrival time of surface reflections from the noise-suppressed data, thereby estimating and correcting flight altitude variations. Next, singular value decomposition based on a distance threshold is applied to suppress surface reflections in the altitude-corrected data. Finally, a time-delay summation imaging method based on refraction point approximation in a two-layer medium model is derived, and the filtered data is focused and imaged using the average flight altitude. This invention achieves accurate detection and location of underground targets.

Description

Technical Field

[0001] This invention belongs to the field of ground penetrating radar technology, specifically relating to a UAV-GPR noise suppression and imaging method and device based on Hankel trajectory matrix decomposition. Background Technology

[0002] Ground penetrating radar (GPR) technology has become an important non-destructive testing method due to its sensitivity to differences in media and its ability to detect both metallic and non-metallic targets. In recent years, with the maturity of unmanned aerial vehicle (UAV) platform technology, mounting GPR systems on UAVs to form UAV-GPR systems has enabled long-range, non-contact, and rapid scanning of hazardous areas, significantly improving operational safety.

[0003] However, in actual flight, drones are susceptible to factors such as airflow and control, making it difficult to maintain absolute stability. This causes dynamic changes in the relative distance between the radar antenna and the ground. This platform jitter introduces two serious problems into the radar echo data: first, it generates strong, non-stationary spatial random noise, masking the effective signal; second, it causes the arrival time of reflected waves from the ground to fluctuate with flight altitude, creating a distorted reflection interface. These problems collectively degrade radar image quality, posing significant challenges to subsequent ground reflection identification, altitude correction, and focused imaging of underground targets.

[0004] Currently, altitude correction for UAV-GPR mainly relies on two types of methods: one is based on additional hardware, such as laser altimeters or differential GPS, to measure the platform's altitude in real time and perform compensation, but this increases system cost, weight, and complexity; the other is based on radar echo signal processing, such as peak detection and edge detection algorithms to automatically determine the ground reflection time. However, existing signal processing methods suffer from a significant drop in positioning accuracy under strong non-stationary noise environments, and generally do not adequately consider the coupling relationship between noise suppression and altitude correction, wavefield separation, and imaging models, resulting in a fragmented processing flow and making it difficult to address the overall challenges of UAV-GPR data processing.

[0005] Therefore, there is an urgent need for a focusing imaging method that can synergistically handle non-stationary noise suppression, robust surface reflection extraction, flight altitude correction, and adapt to dual-layer media, so as to fully realize the potential of the UAV-GPR system in rapid and safe detection. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a UAV-GPR noise suppression and imaging method and apparatus based on Hankel trajectory matrix decomposition. In airborne ground-penetrating radar echo images, it effectively eliminates non-stationary noise, accurately locates the arrival time of surface reflections and completes height correction, thereby achieving focused imaging and precise positioning of underground targets.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition, the method comprising:

[0009] Step 1: Preprocess the collected UAV-GPR radar echo data, construct the Hankel-block-Hankel trajectory matrix based on the preprocessed data, and perform singular value decomposition and reconstruction to obtain noise-suppressed ground-penetrating radar data.

[0010] Step 2: Based on the noise-suppressed ground-penetrating radar data, extract the time of arrival of ground reflection using the derivative threshold reference point method, estimate the UAV's altitude and average flight altitude, and perform altitude correction.

[0011] Step 3: Perform singular value decomposition on the height-corrected ground-penetrating radar data again. Based on the comparison results of the peak positions of each intrinsic sub-image and the preset distance threshold, remove the surface reflection waves to obtain the filtered ground-penetrating radar data.

[0012] Step 4: Based on the time-delay summation imaging method based on the refraction point approximation in the two-layer medium model, the filtered ground-penetrating radar data is focused and imaged using the corrected average flight altitude.

[0013] Furthermore, in step 1, constructing the Hankel-block-Hankel trajectory matrix and performing singular value decomposition and reconstruction includes:

[0014] On the preprocessed radar data matrix, a fixed-size two-dimensional sliding window is slid in unit steps, and the local submatrices covered by each window are rearranged into column vectors in row priority order.

[0015] All the column vectors obtained by sliding are stacked side by side in spatial sliding order to form a trajectory matrix of block Hankel structure;

[0016] Singular value decomposition is performed on the trajectory matrix to obtain a sequence of singular values;

[0017] Based on the singular value sequence, the boundary point between signal and noise is adaptively determined using a local curvature metric. The first few singular values ​​greater than the boundary point are then selected to reconstruct the signal subspace.

[0018] By employing an inverse Hankelization process with subdiagonal averaging, the reconstructed signal subspace is restored to a noise-suppressed data matrix with the same dimension as the original data.

[0019] Furthermore, the adaptive determination of the signal-to-noise boundary point based on the singular value sequence-like local curvature metric includes:

[0020] Calculate the first-order and second-order difference sequences of singular value sequences;

[0021] Based on the first-order difference sequence and the second-order difference sequence, calculate the class local curvature metric value corresponding to each index position. The numerator of the metric value is the square of the second-order difference at that position, and the denominator is the square of the sum of the squares of the constant and the first-order difference.

[0022] A sensitivity factor is set, and the position where the last value of the local curvature metric is greater than the sensitivity factor is determined as the boundary point between the signal subspace and the noise subspace.

[0023] Furthermore, in step 2, extracting the arrival time of surface reflection using the derivative threshold reference point method includes:

[0024] Calculate the discrete derivative sequence of each A-Scan time-domain signal after noise suppression;

[0025] The maximum value of the discrete derivative sequence is multiplied by a preset correction coefficient to obtain the judgment threshold;

[0026] In the time-domain signal, find the position where the first local positive peak exceeds the determination threshold, and mark it as a reference point;

[0027] The nearest local minimum point after the reference point is selected, and its corresponding time sampling point index is the arrival time of the surface reflection.

[0028] Furthermore, in step 3, the method for determining the preset distance threshold is as follows: it is calculated based on the estimated average flight altitude, specifically the electromagnetic wave two-way travel time corresponding to the average flight altitude, plus half of the free space wavelength at the maximum operating frequency.

[0029] Furthermore, in step 4, the refraction point is located on the line connecting the vertical projection point of the radar antenna on the ground surface and the vertical projection point of the underground target on the ground surface; its specific location is determined by the ratio of the relative permittivity of air to the relative permittivity of soil, so that the refraction point is closer to the projection point on the side of the medium with the larger permittivity.

[0030] Furthermore, after performing height correction, step 2 also includes applying a horizontal windowing smoothing filter to the corrected B-Scan image to enhance the horizontal continuity of the image.

[0031] On the other hand, the present invention provides a UAV-GPR noise suppression and imaging device based on Hankel trajectory matrix decomposition, comprising:

[0032] The preprocessing module is used to preprocess the acquired UAV-GPR radar echo data, construct a Hankel-block-Hankel trajectory matrix based on the preprocessed data, and perform singular value decomposition and reconstruction to obtain noise-suppressed ground-penetrating radar data.

[0033] The extraction module is used to extract the time of arrival of ground reflection based on the noise-suppressed ground penetrating radar data using the derivative threshold reference point method, estimate the UAV's altitude and average flight altitude, and perform altitude correction.

[0034] The filtering module is used to perform singular value decomposition on the height-corrected ground-penetrating radar data again. Based on the comparison results of the peak positions of each intrinsic sub-image and the preset distance threshold, the surface reflection waves are removed to obtain the filtered ground-penetrating radar data.

[0035] The imaging module is used to perform focused imaging of filtered ground-penetrating radar data using a time-delay summation imaging method based on the refraction point approximation in a two-layer medium model, utilizing the corrected average flight altitude.

[0036] Thirdly, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition.

[0037] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition.

[0038] The beneficial effects of this invention are as follows:

[0039] Effective suppression of non-stationary noise and preservation of target signals: By constructing a Hankel-block-Hankel trajectory matrix to perform overall analysis of two-dimensional radar data and adopting an adaptive singular value truncation strategy based on a quasi-local curvature metric, it is possible to effectively separate the noise subspace and signal subspace under strong non-stationary noise environment. While suppressing noise, it preserves the structural reflection characteristics of underground targets to the greatest extent, overcoming the problem of the sharp performance degradation of traditional methods under UAV-GPR jitter noise.

[0040] Robust and accurate extraction of surface reflection arrival time: The proposed derivative threshold reference point method, by combining signal derivative characteristics with an adaptive threshold and introducing a judgment rule for the nearest local minimum after the reference point, significantly improves the accuracy and anti-interference ability of surface reflection time positioning, reduces the risk of misjudgment caused by shallow targets or noise, and provides a reliable basis for subsequent height correction.

[0041] Efficient and accurate dual-layer medium imaging: By deriving a time-delay summation imaging formula based on the refraction point approximation, the computational burden of solving complex high-order equations is avoided. Under the premise of ensuring accurate modeling of the air-soil interface refraction effect, the computational efficiency of the imaging algorithm is greatly improved, enabling rapid and accurate focusing and positioning of underground targets.

[0042] Achieving system-level lightweighting and automation: The entire method is implemented using pure software algorithms, eliminating the need for additional hardware modules such as laser altimeters for height measurement. This reduces the load, cost, and operational complexity of the UAV-GPR system, promoting its development towards lightweighting and intelligence. At the same time, it achieves fully automated processing from raw data to target imaging through a complete processing chain. Attached Figure Description

[0043] Figure 1 This is a flowchart of the UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition of the present invention;

[0044] Figure 2 A schematic diagram illustrating the detection and electromagnetic wave propagation of an airborne ground-penetrating radar system.

[0045] Figure 3 This is a schematic diagram of the drone's flight process;

[0046] Figure 4 A schematic diagram of the image after preprocessing the raw data collected by the UAV-borne ground-penetrating radar system;

[0047] Figure 5 This is a schematic diagram of ground-penetrating radar data processed according to the method of the present invention;

[0048] Figure 6 This is a schematic diagram of the time-domain signal surface reflection arrival time positioning process proposed in this invention;

[0049] Figure 7(a) shows the imaging results of the method of the present invention;

[0050] Figure 7(b) shows the imaging results of the peak method;

[0051] Figure 7(c) shows the imaging results of the Sobel operator method;

[0052] Figure 7 (d) shows the imaging results of the local energy curve method. Detailed Implementation

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0054] like Figure 1 As shown, this invention provides a UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition, comprising:

[0055] Step 1: Preprocess the collected UAV-GPR radar echo data, construct the Hankel-block-Hankel trajectory matrix based on the preprocessed data, and perform singular value decomposition and reconstruction to obtain noise-suppressed ground-penetrating radar data.

[0056] Step 2: Based on the noise-suppressed ground-penetrating radar data, extract the time of arrival of ground reflection using the derivative threshold reference point method, estimate the UAV's altitude and average flight altitude, and perform altitude correction.

[0057] Step 3: Perform singular value decomposition on the height-corrected ground-penetrating radar data again. Based on the comparison results of the peak positions of each intrinsic sub-image and the preset distance threshold, remove the surface reflection waves to obtain the filtered ground-penetrating radar data.

[0058] Step 4: Based on the time-delay summation imaging method based on the refraction point approximation in the two-layer medium model, the filtered ground-penetrating radar data is focused and imaged using the corrected average flight altitude.

[0059] Specifically, step 1 includes:

[0060] In this invention, the UAV-GPR collects A-Scan data along a single flight survey line at a certain altitude above the ground. Through these continuously collected A-Scan data, B-Scan ground-penetrating radar data for that flight survey line is obtained. A schematic diagram of the detection method is shown below. Figure 2 As shown.

[0061] Preprocessing of the acquired ground-penetrating radar data includes: background clutter removal, A-Scan peak normalization, and time window truncation.

[0062] In ground-penetrating radar (GPR) data processing, due to the inherent characteristics of the radar system, background clutter in B-Scan images typically exhibits a constant horizontal offset. A simple and effective suppression method is to calculate the spatial average along the flight path for each time sampling point and then subtract it. Due to factors such as antenna transmit power and survey line altitude variations, the absolute amplitudes between different A-Scans often differ significantly. Directly using the original amplitudes can affect the reliability of subsequent threshold detection; therefore, peak value normalization is necessary to achieve amplitude consistency. The purpose of time window truncation is to extract the effective time period containing target information and remove signals irrelevant to the target, thereby reducing data size and improving the signal-to-noise ratio.

[0063] Assume the preprocessed ground-penetrating radar data is of size [size missing]. The matrix is ​​represented as the following matrix. form:

[0064] (1)

[0065] in, This indicates the number of sampling points in the time series after the time window is truncated. This represents the total number of channels acquired, and each element in the matrix represents the amplitude of the radar data at the corresponding location.

[0066] Define a size of A two-dimensional window matrix is ​​used to extract local information from radar data. This window matrix is ​​arranged from left to right and top to bottom, with a step size of one pixel within the data matrix. Slide the radar data matrix to the center. As the starting point of the window matrix, the number of windows moved horizontally is... The number of windows that move vertically is The total number of sliding windows is Among them, location The data within the window can be represented as a local submatrix. :

[0067] (2)

[0068] Each local submatrix The elements are rearranged horizontally in a fixed order, thus mapping the local submatrix to column vectors. :

[0069] (3)

[0070] The column vectors after mapping all local submatrices are stacked side-by-side according to the window sliding order, resulting in a structure of size [size missing]. trajectory matrix :

[0071] (4)

[0072] The constructed trajectory matrix exhibits a Hankel-block-Hankel (HBH) structure, which can be viewed as a series of sequentially arranged block matrices. Composition, block matrix It also possesses a Hankel structure. Specifically, the trajectory matrix... It can be represented as:

[0073] (5)

[0074] Wherein, block matrix Represented as:

[0075] (6)

[0076] Applying singular value decomposition to the Hankel trajectory matrix decomposes it into subspaces with different components:

[0077] (7)

[0078] in, and It is the size of and unitary matrix, It is the size of A diagonal matrix whose elements are matrices. singular values, and Each is a matrix and The Column vectors diagonal matrix The diagonal elements, Let be the rank of the trajectory matrix.

[0079] Based on the compositional characteristics of the UAV-GPR profile, the singular spectrum formed by the diagonal elements of the singular value matrix can be considered to consist of two parts: noise-free effective signal and non-stationary noise. The first few terms corresponding to larger singular values ​​usually represent the dominant information of the data, while the terms corresponding to smaller singular values ​​contain weak or random components submerged in noise. The terms are considered as signal subspaces and used for reconstruction:

[0080] (8)

[0081] For the reconstructed signal subspace signal, the threshold The choice of [a specific element] determines the effectiveness of the filtering. When the value is too high, it may lead to excessive filtering of useful information related to the target. When the value is too low, noise cannot be sufficiently suppressed. This invention proposes a quasi-local curvature metric that adaptively determines the optimal threshold. The choice is made to strike a balance between suppressing noise and preserving the target signal.

[0082] Extracting the diagonal matrix The singular values, arranged in order, form a one-dimensional discrete sequence. First-order and second-order differences are defined as follows:

[0083] (9)

[0084] (10)

[0085] For each index Computational Local Curvature Measurement Index :

[0086] (11)

[0087] Using the squared value at the numerator of a local curvature metric can significantly amplify changes in the second-order difference and eliminate the influence of the sign, thus more robustly highlighting locations where local curvature increases significantly. The larger the value, the more significant the inflection point in the singular spectrum at that position.

[0088] make Optimal cutoff threshold The selection rules are as follows:

[0089] (12)

[0090] in, is a conservative sensitivity factor used to identify bends with statistical or physical significance, and z is an intermediate parameter.

[0091] Reconstructed signal subspace signal The data matrix is ​​restored to its original form through the inverse Hankelization process. The matrix components corresponding to the size. First, the reconstructed signal subspace signal... The block matrices are averaged along their sub-diagonal sides to achieve a Hankel structure. Sub-diagonal averaging is then performed at the inter-block level to obtain matrices with a Hankel-block-Hankel structure. Since these matrices have a unique correspondence with the original matrices, they can be de-Hankelized to the original data dimension, i.e., the noise-suppressed radar data matrix. as follows:

[0092] (13)

[0093] In this matrix, each element represents the amplitude of the corresponding location in the radar image after noise suppression.

[0094] Step 2 includes:

[0095] The arrival time of ground reflection is extracted using the A-Scan time-domain waveform after non-stationary noise suppression. Let the single-channel discrete time-domain signal be... The sampling interval is Sampling points To characterize the instantaneous features of a signal, the discrete derivative is calculated. :

[0096] (14)

[0097] The maximum value of the discrete derivative sequence As the initial reference quantity, that is:

[0098] (15)

[0099] Introducing correction coefficients The initial reference value is reduced or enlarged to obtain the judgment threshold. :

[0100] (16)

[0101] in, Determined by field or calibration data, it is used to compensate for the systematic impact of system response and scene changes on the derivative amplitude, thereby improving the robustness of threshold determination under different operating conditions.

[0102] To stably indicate the location of surface reflections, a judgment threshold is obtained. Then, search for discrete signals. The first local positive peak value in the sampled data exceeds the judgment threshold, and its corresponding sampling point index is determined. This is recorded as the reference point, i.e.:

[0103] (17)

[0104] Select distance reference point The location of the nearest local minimum Mathematically, the index of the time sampling point corresponding to the arrival time of surface reflection can be defined as:

[0105] (18)

[0106] Indexed by arrival time sampling point Calculate arrival time When converting to altitude, the speed of electromagnetic waves in the air should be considered. and round-trip time latency characteristics, drone altitude The average flight altitude can be calculated using the following formula:

[0107] (19)

[0108] (20)

[0109] in, This indicates the total number of channels collected. This represents the flight altitude of the corresponding lane. After obtaining the arrival times of ground reflection for all A-scans, the time delay differences caused by changes in flight altitude are corrected. Specifically, firstly, the arrival time of ground reflection for each A-scan is compared with the average arrival time of all lanes, and its relative offset is calculated. This offset is then considered as the manifestation of flight altitude changes on the time axis. Subsequently, based on this offset information, each A-scan waveform is translated on the time axis so that the position of the ground reflection is relative to the average flight altitude. Alignment is achieved to achieve height correction. To further improve the overall quality and lateral continuity of the corrected B-Scan image, a windowed smoothing filter is applied to the corrected image to effectively smooth the lateral distribution of surface reflection interfaces and underground target reflections.

[0110] Step 3 includes:

[0111] First, the ground-penetrating radar data after height correction obtained in step 2 is subjected to singular value decomposition according to formula (7) to obtain several intrinsic sub-images and corresponding singular values.

[0112] Then, the intrinsic sub-images are classified according to their amplitude distribution spatial location. This is done by calculating the location of the global maximum value for each intrinsic sub-image and comparing this location with a pre-determined distance threshold. This distance threshold is used to approximate the location of the air-to-surface interface; intrinsic sub-images with peak locations within this threshold are considered primarily composed of surface reflections and are discarded. Intrinsic sub-images with peak locations outside this threshold are retained as components that may contain information about subsurface targets.

[0113] The distance threshold is determined based on the average flight altitude estimated from radar data in step 2. , specifically set ,in The wavelength in free space at the maximum operating frequency.

[0114] Finally, the retained intrinsic sub-images and their singular values ​​are used for data reconstruction to obtain the filtered ground-penetrating radar data matrix. The reconstructed data matrix reduces the energy of surface reflected waves while retaining the scattered energy of underground targets.

[0115] Step 4 includes:

[0116] UAV-GPR operates between two media, air and ground. Electromagnetic waves need to pass through the ground surface to enter the underground medium. The dielectric constant, as a direct factor affecting the propagation speed and path of electromagnetic waves, exhibits abrupt changes in dielectric constant between air and the underground medium, leading to refraction of the electromagnetic waves. The electromagnetic wave propagation model is as follows: Figure 2 As shown.

[0117] Traditional methods derive information about the refraction point based on Snell's law of refraction. A quartic equation in one variable was proposed, and the equation was solved using numerical optimization methods. However, solving this equation requires a large amount of computation, which seriously affects imaging efficiency. Considering that the refraction point is located at the intersection of the electromagnetic wave and the Earth's surface during rectilinear propagation... and the vertical projection of the target on the ground Between, for the refraction point Make the following approximation:

[0118] (twenty one)

[0119] in, and Let be the relative permittivity of air and soil, respectively. The refraction point is obtained using an approximation method. Then, the location of the radar antenna in the delay summation algorithm. To the imaging point Delay between for:

[0120] (twenty two)

[0121] in, The average flight altitude of the drone obtained in step 2. Represents the speed of light. Indicates the location of the radar antenna. This represents the location of the imaging point.

[0122] The filtered ground-penetrating radar data obtained in step 3 is focused and imaged according to formula (22).

[0123] To verify the feasibility of the UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition proposed in this invention, a flight experiment was conducted in an open area of ​​a UAV flight training base. All flight experiments were carried out in a flat environment, and the experimental results were analyzed.

[0124] In this experiment, two landmine models were buried along the same survey line. One landmine model with a radius of 13cm was buried at a depth of 5cm, and the other landmine model with a radius of 7cm was buried at a depth of 3cm. The horizontal distance between the targets was approximately 1.5m. The UAV flew at a constant speed of 0.5m / s along the survey line at a height of approximately 1.5m. Figure 3 This is a schematic diagram of the drone's flight process.

[0125] Figure 4 The images obtained from the raw data acquired by the UAV-borne ground-penetrating radar system, after preprocessing, clearly show air-to-surface reflected waves between approximately 8 and 10 ns, with the surface reflected waves exhibiting a distinct undulating characteristic. Significant differences in the energy of the echo signals received at different observation locations further exacerbate the non-stationary noise due to these spatial variations in height and amplitude. Two independent hyperbolas appear within a 9-10 ns time window, corresponding to the responses of two underground targets.

[0126] To address the problem of non-stationary noise, the non-stationary noise suppression method proposed in this invention is employed. Figure 5 This is ground-penetrating radar data processed according to the method of the present invention. Compared with noisy radar images, the non-stationary noise in the processed radar images is significantly reduced, the background texture tends to be smoother, and the structural reflection features of the ground surface and targets are better preserved.

[0127] Figure 6 This is a schematic diagram illustrating the extraction of the ground reflection arrival time using the derivative threshold reference point method proposed in this invention, based on one of the A-Scan time-domain waveforms after non-stationary noise suppression. By utilizing the proposed ground reflection arrival time localization method and finding the nearest local minimum after the reference point, the ground reflection arrival time can be accurately located, providing detailed and reliable data support for subsequent calculations of UAV average flight altitude information and altitude correction.

[0128] Figures 7(a)-7(d) The results are obtained by summing the delays after locating the arrival time of surface reflection using the algorithm, peak method, Sobel operator edge detection method, and local energy curve method proposed in this invention. Figure 7(a) shows that the focused image of the proposed method can clearly show the high-energy focal points of the two targets, while the focused images of Figures 7(b), 7(c), and 7(d) show obvious energy diffusion and false speckle, making it difficult to effectively distinguish the targets.

[0129] On the other hand, the present invention provides a UAV-GPR noise suppression and imaging device based on Hankel trajectory matrix decomposition, which includes modules capable of implementing the steps of the aforementioned method, specifically including:

[0130] The preprocessing module is used to preprocess the acquired UAV-GPR radar echo data, construct a Hankel-block-Hankel trajectory matrix based on the preprocessed data, and perform singular value decomposition and reconstruction to obtain noise-suppressed ground-penetrating radar data.

[0131] The extraction module is used to extract the time of arrival of ground reflection based on the noise-suppressed ground penetrating radar data using the derivative threshold reference point method, estimate the UAV's altitude and average flight altitude, and perform altitude correction.

[0132] The filtering module is used to perform singular value decomposition on the height-corrected ground-penetrating radar data again. Based on the comparison results of the peak positions of each intrinsic sub-image and the preset distance threshold, the surface reflection waves are removed to obtain the filtered ground-penetrating radar data.

[0133] The imaging module is used to perform focused imaging of filtered ground-penetrating radar data using a time-delay summation imaging method based on the refraction point approximation in a two-layer medium model, utilizing the corrected average flight altitude.

[0134] Thirdly, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition.

[0135] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition.

[0136] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition, characterized in that, The method includes: Step 1: Preprocess the collected UAV-GPR radar echo data, construct the Hankel-block-Hankel trajectory matrix based on the preprocessed data, and perform singular value decomposition and reconstruction to obtain noise-suppressed ground-penetrating radar data. Step 2: Based on the noise-suppressed ground-penetrating radar data, extract the time of arrival of ground reflection using the derivative threshold reference point method, estimate the UAV's altitude and average flight altitude, and perform altitude correction. Step 3: Perform singular value decomposition on the height-corrected ground-penetrating radar data again. Based on the comparison results of the peak positions of each intrinsic sub-image and the preset distance threshold, remove the surface reflection waves to obtain the filtered ground-penetrating radar data. Step 4: Based on the time-delay summation imaging method based on the refraction point approximation in the two-layer medium model, the filtered ground-penetrating radar data is focused and imaged using the corrected average flight altitude.

2. The UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition according to claim 1, characterized in that, Step 1, which involves constructing the Hankel-block-Hankel trajectory matrix and performing singular value decomposition and reconstruction, includes: On the preprocessed radar data matrix, a fixed-size two-dimensional sliding window is slid in unit steps, and the local submatrices covered by each window are rearranged into column vectors in row priority order. All the column vectors obtained by sliding are stacked side by side in spatial sliding order to form a trajectory matrix of block Hankel structure; Singular value decomposition is performed on the trajectory matrix to obtain a sequence of singular values; Based on the singular value sequence, the boundary point between signal and noise is adaptively determined using a local curvature metric. The first few singular values ​​greater than the boundary point are then selected to reconstruct the signal subspace. By employing an inverse Hankelization process with subdiagonal averaging, the reconstructed signal subspace is restored to a noise-suppressed data matrix with the same dimension as the original data.

3. The UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition according to claim 2, characterized in that, The adaptive determination of the signal-to-noise boundary point based on the singular value sequence-like local curvature metric includes: Calculate the first-order and second-order difference sequences of singular value sequences; Based on the first-order difference sequence and the second-order difference sequence, calculate the class local curvature metric value corresponding to each index position. The numerator of the metric value is the square of the second-order difference at that position, and the denominator is the square of the sum of the squares of the constant and the first-order difference. A sensitivity factor is set, and the position where the last value of the local curvature metric is greater than the sensitivity factor is determined as the boundary point between the signal subspace and the noise subspace.

4. The UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition according to claim 1, characterized in that, In step 2, extracting the arrival time of surface reflection using the derivative threshold reference point method includes: Calculate the discrete derivative sequence of each A-Scan time-domain signal after noise suppression; The maximum value of the discrete derivative sequence is multiplied by a preset correction coefficient to obtain the judgment threshold; In the time-domain signal, find the position where the first local positive peak exceeds the determination threshold, and mark it as a reference point; The nearest local minimum point after the reference point is selected, and its corresponding time sampling point index is the arrival time of the surface reflection.

5. The UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition according to claim 1, characterized in that, In step 3, the method for determining the preset distance threshold is as follows: it is calculated based on the estimated average flight altitude, and the specific value is the electromagnetic wave two-way travel time corresponding to the average flight altitude, plus half of the free space wavelength at the maximum operating frequency.

6. The UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition according to claim 1, characterized in that, In step 4, the refraction point is located on the line connecting the vertical projection point of the radar antenna on the ground surface and the vertical projection point of the underground target on the ground surface; its specific location is determined by the ratio of the relative permittivity of air to the relative permittivity of soil, so that the refraction point is closer to the projection point on the side of the medium with the larger permittivity.

7. The UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition according to claim 1, characterized in that, After performing height correction, step 2 also includes applying a horizontal windowing smoothing filter to the corrected B-Scan image to enhance the horizontal continuity of the image.

8. A UAV-GPR noise suppression and imaging device based on Hankel trajectory matrix decomposition, characterized in that, include: The preprocessing module is used to preprocess the acquired UAV-GPR radar echo data, construct a Hankel-block-Hankel trajectory matrix based on the preprocessed data, and perform singular value decomposition and reconstruction to obtain noise-suppressed ground-penetrating radar data. The extraction module is used to extract the time of arrival of ground reflection based on the noise-suppressed ground penetrating radar data using the derivative threshold reference point method, estimate the UAV's altitude and average flight altitude, and perform altitude correction. The filtering module is used to perform singular value decomposition on the height-corrected ground-penetrating radar data again. Based on the comparison results of the peak positions of each intrinsic sub-image and the preset distance threshold, the surface reflection waves are removed to obtain the filtered ground-penetrating radar data. The imaging module is used to perform focused imaging of filtered ground-penetrating radar data using a time-delay summation imaging method based on the refraction point approximation in a two-layer medium model, utilizing the corrected average flight altitude.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the UAV-GPR noise suppression and imaging method based on Hankel trajectory matrix decomposition as described in any one of claims 1-7.

Images