A coal mine underground strong clutter suppression method based on space-time-frequency joint sparse reconstruction

Abstract

The application discloses a coal mine underground strong clutter suppression method based on space-time-frequency joint sparse reconstruction, and relates to the technical field of radar signal processing. The method comprises the following steps: S1, receiving a baseband signal, forming a space-time two-dimensional data matrix, and then constructing a space-time-frequency three-dimensional data cube of a radar echo; S2, vectorizing the space-time-frequency three-dimensional data cube to obtain an observation vector, and constructing a joint sparse dictionary; S3, based on the joint sparse dictionary and the observation vector, a joint sparse optimization model is established; S4, a sparse reconstruction algorithm is used to solve the established joint sparse optimization model, and sparse coefficients of a target signal are obtained, and a target signal after clutter suppression is reconstructed; the application utilizes the difference between the joint sparsity of the target in the space-time-frequency three domains and the local sparsity of the clutter, realizes deep clutter suppression, and even under the condition that the input signal-to-clutter ratio is as low as-5 dB, the output signal-to-clutter ratio improvement can reach 23.9 dB or more.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, specifically to a method for suppressing strong clutter in coal mines based on space-time-frequency joint sparse reconstruction. Background Technology

[0002] Coal mine safety production heavily relies on the accurate detection of hidden hazardous geological bodies (such as goafs and water-bearing structures) and workers underground. Through-layer detection radar shows promise in this field due to its non-contact and high-resolution characteristics. However, coal mine roadways are typical environments with strong clutter. Metal supports, conveyors, cables, and other facilities strongly reflect electromagnetic waves, creating clutter intensity several orders of magnitude higher than the echoes from targets (such as people or goaf interfaces). This strong clutter background severely obscures weak target signals, leading to a sharp decline in the performance of traditional radar detection methods.

[0003] Commonly used clutter suppression methods include: mean cancellation, which relies on the characteristic that clutter energy is concentrated near zero frequency in the time domain (Doppler domain). It suppresses static clutter by calculating the average value of adjacent pulse echoes and performing a subtraction operation. However, its ability to distinguish slow-moving targets or distributed clutter with similar Doppler characteristics is limited, and it easily causes target signal distortion. Principal component analysis (PCA), as a blind source separation technique, can suppress major clutter components, but it cannot effectively utilize the joint information in the spatial and frequency domains, resulting in incomplete clutter suppression and significant target signal energy loss (typically above 15%). Space-time adaptive processing (STAP), while effective in airborne radar, heavily relies on a large number of training samples satisfying the independent and identically distributed condition to accurately estimate clutter statistical characteristics, which is difficult to achieve in complex and time-varying coal mine roadways. Furthermore, STAP has high computational complexity, posing a challenge to the real-time processing capabilities of underground equipment.

[0004] In recent years, compressed sensing and sparse reconstruction theory have shown that if a signal is sparse in a certain transform domain (dictionary), the original signal can be recovered with high probability from a small amount of observation data. Most existing clutter suppression methods based on sparse reconstruction are limited to processing in a single domain (such as spatial or temporal domain) or two domains (such as spatial-temporal domain), failing to fully utilize the joint sparsity characteristics of the target in all three domains. Therefore, when facing extremely strong clutter in coal mines, their clutter suppression depth and target fidelity remain unsatisfactory. Thus, there is an urgent need for a novel signal processing method that can adapt to the non-uniform, highly cluttered environment of coal mines and achieve deep clutter suppression and high-fidelity target signal reconstruction. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for suppressing strong clutter in coal mines based on space-time-frequency joint sparse reconstruction.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] This application discloses a method for suppressing strong clutter in coal mines based on joint space-time-frequency sparse reconstruction, including the following steps:

[0008] S1. The baseband signal is received by a multi-input multi-output radar linear array system deployed in the coal mine roadway, forming a space-time two-dimensional data matrix, and then a space-time-frequency three-dimensional data cube of the radar echo is constructed.

[0009] S2. Vectorize the space-time-frequency three-dimensional data cube to obtain the observation vector, and construct a joint sparse dictionary, which includes a target sub-dictionary and a clutter sub-dictionary.

[0010] S3. Establish a joint sparse optimization model based on the joint sparse dictionary and observation vector;

[0011] S4. The sparse reconstruction algorithm is used to solve the established joint sparse optimization model to obtain the sparse coefficients of the target signal and reconstruct the target signal after clutter suppression.

[0012] Preferably, the construction of the space-time-frequency three-dimensional data cube of the radar echo in step S1 specifically includes: performing a range-dimensional Fourier transform on the array element-pulse echo in the space-time two-dimensional data matrix, mapping it to the range-frequency domain, thereby constructing the space-time-frequency three-dimensional data cube of the radar echo. , ,in Indicates the target component, Indicates clutter components, This represents the noise component.

[0013] Preferably, step S2 specifically includes: converting the space-time-frequency three-dimensional data cube Vectorize by column to obtain the observation vector. , Construct a joint sparse dictionary , ,in Represents the target sub-dictionary. This represents a dictionary of miscellaneous sub-dictionaries.

[0014] Preferably, the target sub-dictionary atoms From azimuth Doppler frequency and distance frequency The Kronecker product is constructed, i.e. , used to characterize the joint sparsity of the target in the space-time-frequency domains, where a represents the spatial steering vector, b represents the time steering vector, and d represents the frequency response vector. This represents the Kronecker product operation.

[0015] Preferably, for clutter sub-dictionaries Static clutter in the atom A typical sparse structure used to capture clutter is constructed as follows: ,in Indicates length as the number of pulses A vector of all 1s.

[0016] Preferably, step S3 specifically includes: establishing a joint sparse optimization model based on the joint sparse dictionary and the observation vector. ,in Represents a sparse coefficient vector. and Represents the regularization parameter. Represents the set of target atoms. Represents the target coefficient. Represents the clutter coefficient. Describing the L2 norm, Describing the L1 norm, This represents the estimation of sparse coefficients. Indicates signal reconstruction, Indicates that the target coefficient vector belongs to the first... The coefficient of the group.

[0017] Preferably, step S4 uses an improved joint orthogonal matching pursuit algorithm to iteratively solve the joint sparse optimization model, specifically including the following steps:

[0018] S41. Set initial residuals , Initialize the target support set clutter support set The target support set clutter support set Set the maximum number of iterations for an empty set. and residual threshold ;

[0019] S42, in the target sub-dictionary Find the set of atoms most relevant to the current residual in the dictionary of clutter sub-dictionaries. Find the single atom most relevant to the current residual; then add the index of the selected atom group and the index of the atom to the target support set. clutter support set Based on the current joint support set Update the sparsity coefficients using the least squares method Through formula For residuals Update if residual satisfy Or the number of iterations reaches the maximum number of iterations. If the condition is met, stop iterating; otherwise, continue iterating. This represents the submatrix composed of dictionary atoms selected from the current joint support set;

[0020] S43. Output Sparsity Coefficient Estimation The target signal was then reconstructed. , And restore it to a three-dimensional data cube This refers to the target signal after clutter suppression.

[0021] The beneficial effects of this invention are:

[0022] 1) This application makes full use of the difference between the joint sparsity of the target in the space-time-frequency domain and the local sparsity of clutter to achieve deep clutter suppression. Even when the input signal-to-clutter ratio is as low as -5 dB, the output signal-to-clutter ratio improvement can reach more than 23.9 dB.

[0023] 2) While suppressing clutter, this application can retain the target signal energy to the maximum extent and control the target energy loss to within 3%, which is far superior to PCA (approximately 18.7% loss) and STAP (approximately 12.5% ​​loss), providing a high-quality data foundation for subsequent target identification and parameter estimation.

[0024] 3) This invention does not rely on the accurate estimation of clutter statistical characteristics and has good robustness to complex clutter environments that are non-uniform and non-Gaussian in underground coal mines, overcoming the shortcomings of existing technologies that depend on uniform samples.

[0025] 4) This application achieves intelligent signal processing by constructing a joint dictionary and designing structural sparse constraints to separate the target from clutter, rather than through simple filtering. Attached Figure Description

[0026] Figure 1 This is a schematic diagram illustrating the steps of a coal mine underground strong clutter suppression method based on space-time-frequency joint sparse reconstruction according to an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram comparing the distance-Doppler spectrum of an embodiment of the present invention with that of the prior art at an input signal-to-noise ratio of -5 dB;

[0028] Figure 3This is a comparative schematic diagram of the embodiments of the present invention and the prior art, wherein (a) is a comparative schematic diagram of the performance improvement of signal-to-clutter ratio in deep detection, and (b) is a comparative schematic diagram of the detection capability of deep geological structures. Detailed Implementation

[0029] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] This application discloses a method for suppressing strong clutter in coal mines based on joint space-time-frequency sparse reconstruction, aiming to solve the problem of weak targets being submerged by strong clutter under extremely low signal-to-clutter ratio conditions, and achieving excellent clutter suppression performance and target signal fidelity. A schematic diagram of the method's steps is shown below. Figure 1 As shown, the specific steps include:

[0031] S1. The baseband signal is received by a multi-input multi-output radar linear array system deployed in the coal mine roadway, forming a space-time two-dimensional data matrix, and then a space-time-frequency three-dimensional data cube of the radar echo is constructed.

[0032] S2. Vectorize the space-time-frequency three-dimensional data cube to obtain the observation vector, and construct a joint sparse dictionary, which includes a target sub-dictionary and a clutter sub-dictionary.

[0033] S3. Establish a joint sparse optimization model based on the joint sparse dictionary and observation vector;

[0034] S4. The sparse reconstruction algorithm is used to solve the established joint sparse optimization model to obtain the sparse coefficients of the target signal and reconstruct the target signal after clutter suppression.

[0035] Specifically, the construction of the space-time-frequency three-dimensional data cube of the radar echo in step S1 includes: performing a range-dimensional (fast time-dimensional) Fourier transform on the array element-pulse echo in the space-time two-dimensional data matrix, mapping it to the range-frequency domain, thereby constructing the space-time-frequency three-dimensional data cube of the radar echo. , ,in Indicates the target component, Indicates clutter components, This represents the noise component.

[0036] Specifically, step S2 includes: converting the space-time-frequency three-dimensional data cube Vectorize by column to obtain the observation vector. , Construct a joint sparse dictionary , ,in Represents the target sub-dictionary. This represents a dictionary of miscellaneous sub-dictionaries.

[0037] Specifically, the target sub-dictionary atoms Discretized spatial guidance vector (azimuth angle) Time-domain steering vector (Doppler frequency) ) and frequency domain response vector (distance frequency) The Kronecker product of ) is constructed, i.e. , used to characterize the joint sparsity of the target in the space-time-frequency domains, where a represents the spatial steering vector, b represents the time steering vector, and d represents the frequency response vector. This represents the Kronecker product operation, used to combine vectors from three fields into a high-dimensional joint atom.

[0038] Specifically, for the clutter sub-dictionary Static clutter in the atom A typical sparse structure used to capture clutter is constructed as follows: ,in Indicates length as the number of pulses A vector of all 1s is used to represent its characteristic of being undulating in the time domain.

[0039] Specifically, step S3 includes: establishing a joint sparse optimization model based on the joint sparse dictionary and the observation vector. ,in Represents a sparse coefficient vector. and This represents the regularization parameter, used to balance the weights of each item. Represents the set of target atoms. Represents the target coefficient. Represents the clutter coefficient. Describing the L2 norm, Describing the L1 norm, This represents the estimate of the sparse coefficients; where For data fidelity items, To apply to the target coefficient The mixed L2 norm regularization term is used to promote the group sparsity of the target atom, which fits the joint sparsity mode of the target in the three domains. To apply clutter coefficient The L1 norm regularization term is used to promote individual sparsity of clutter atoms. Indicates signal reconstruction, Indicates that the target coefficient vector belongs to the first... The coefficient of the group; This structured sparsity is a key design feature that distinguishes this invention from traditional sparse methods by utilizing prior knowledge of the target (joint sparsity). It enables the model to intelligently distinguish between "multi-domain responses caused by a single target" (group sparsity) and "isolated responses caused by clutter" (individual sparsity).

[0040] Specifically, step S4 uses an improved joint orthogonal matching pursuit algorithm to iteratively solve the joint sparse optimization model, which includes the following steps:

[0041] S41. Set initial residuals , Initialize the target support set clutter support set The target support set clutter support set Set the maximum number of iterations for an empty set. and residual threshold ;

[0042] S42, in the target sub-dictionary Find the set of atoms most relevant to the current residual in the dictionary of clutter sub-dictionaries. Find the single atom most relevant to the current residual; then add the index of the selected atom group and the index of the atom to the target support set. clutter support set Based on the current joint support set Update the sparsity coefficients using the least squares method Through formula For residuals Update if residual satisfy Or the number of iterations reaches the maximum number of iterations. If the condition is met, stop iterating; otherwise, continue iterating. This represents the submatrix composed of dictionary atoms selected from the current joint support set;

[0043] S43. Output Sparsity Coefficient Estimation The target signal was then reconstructed. , And restore it to a three-dimensional data cube This refers to the target signal after clutter suppression.

[0044] For example, this application is implemented based on the application scenario of underground personnel detection in coal mines:

[0045] A 16-element uniform linear array is used, with a pulse count of The distance-dimensional FFT point count L=64; a strong scattering metal support (static clutter) is set at 50 meters in the tunnel, and a point target (personnel) moving radially at 0.5 m / s is set at 45 meters. By adjusting the target scattering coefficient, the initial signal-to-clutter ratio is made to -5 dB; in this embodiment, the maximum number of iterations is... Set to 50, residual threshold The distance-Doppler spectrum comparison diagram between this application and the prior art at an input signal-to-noise ratio of -5 dB is shown below. Figure 2 As shown; a comparative diagram of this application and the prior art is shown below. Figure 3 As shown in the figure, (a) is a schematic diagram comparing the signal-to-noise ratio improvement performance of deep detection, and (b) is a schematic diagram comparing the detection capabilities of deep geological structures. As can be seen from the figure, the output signal-to-noise ratio after processing in this application is improved to 18.9 dB, the signal-to-noise ratio improvement is 23.9 dB, and the target energy loss is only 2.4%, which is significantly better than the existing technology.

[0046] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for suppressing strong clutter in coal mines based on joint space-time-frequency sparse reconstruction, characterized in that, Includes the following steps: S1. The baseband signal is received by a multi-input multi-output radar linear array system deployed in the coal mine roadway, forming a space-time two-dimensional data matrix, and then a space-time-frequency three-dimensional data cube of the radar echo is constructed. S2. Vectorize the space-time-frequency three-dimensional data cube to obtain the observation vector, and construct a joint sparse dictionary, which includes a target sub-dictionary and a clutter sub-dictionary. S3. Based on the joint sparse dictionary and observation vector, establish a joint sparse optimization model; that is... ,in Represents a sparse coefficient vector. and Represents the regularization parameter. Represents the set of target atoms. Represents the target coefficient. Represents the clutter coefficient. Let L2 norm be , L1 norm be , and represent the sparse coefficient estimate. Indicates signal reconstruction, Indicates that the target coefficient vector belongs to the first... The coefficient of the group, Represents the observation vector; S4. The sparse reconstruction algorithm is used to solve the established joint sparse optimization model to obtain the sparse coefficients of the target signal and reconstruct the target signal after clutter suppression.

2. The method for suppressing strong clutter in coal mines based on space-time-frequency joint sparse reconstruction according to claim 1, characterized in that, The construction of the space-time-frequency three-dimensional data cube of radar echoes in step S1 specifically includes: performing a range-dimensional Fourier transform on the array element-pulse echo in the space-time two-dimensional data matrix, mapping it to the range-frequency domain, thereby constructing the space-time-frequency three-dimensional data cube of radar echoes. , ,in Indicates the target component, Indicates clutter components, This represents the noise component.

3. The method for suppressing strong clutter in coal mines based on space-time-frequency joint sparse reconstruction according to claim 2, characterized in that, Step S2 specifically includes: converting the space-time-frequency three-dimensional data cube Vectorize by column to obtain the observation vector. , Construct a joint sparse dictionary , ,in Represents the target sub-dictionary. This represents a dictionary of miscellaneous sub-dictionaries.

4. The method for suppressing strong clutter in coal mines based on joint space-time-frequency sparse reconstruction according to claim 3, characterized in that: The target sub-dictionary atoms From azimuth Doppler frequency and distance frequency The Kronecker product is constructed, i.e. , used to characterize the joint sparsity of the target in the space-time-frequency domains, where a represents the spatial steering vector, b represents the time steering vector, and d represents the frequency response vector. This represents the Kronecker product operation.

5. A method for suppressing strong clutter in coal mines based on space-time-frequency joint sparse reconstruction according to claim 4, characterized in that, For clutter dictionary Static clutter in the atom A typical sparse structure used to capture clutter is constructed as follows: ,in Indicates length as the number of pulses A vector of all 1s.

6. The method for suppressing strong clutter in coal mines based on space-time-frequency joint sparse reconstruction according to claim 5, characterized in that, Step S4 uses an improved joint orthogonal matching pursuit algorithm to iteratively solve the joint sparse optimization model, specifically including the following steps: S41. Set initial residuals , Initialize the target support set clutter support set The target support set clutter support set Set the maximum number of iterations for an empty set. and residual threshold ; S42, in the target sub-dictionary Find the set of atoms most relevant to the current residual in the dictionary of clutter sub-dictionaries. Find the single atom most relevant to the current residual; then add the index of the selected atom group and the index of the atom to the target support set. clutter support set Based on the current joint support set Update the sparsity coefficients using the least squares method Through formula For residuals Update if residual satisfy Or the number of iterations reaches the maximum number of iterations. If the condition is met, the iteration stops; otherwise, the iteration continues; represents the submatrix formed by the dictionary atoms selected from the current joint support set; S43. Output Sparsity Coefficient Estimation The target signal was then reconstructed. , And restore it to a three-dimensional data cube This refers to the target signal after clutter suppression.

Images