Echo data sparse acquisition method for microwave near-field three-dimensional imaging

Abstract

The application discloses a kind of for microwave near-field three-dimensional imaging echo data sparse acquisition method, traditional plane scanning area is divided into several sub-regions and obtains initial sampling data, establishes observation model in each sub-region, the uncertainty of new input point sampling data is predicted by observation model, each sub-region extracts the new input point of maximum uncertainty to constitute sparse sampling point;Then use genetic algorithm to plan optimal scanning path, sparse sampling data is collected according to the planning path and is incorporated into initial sampling data, after updating model, the next group of high-uncertainty prediction points is iteratively predicted;Continue iteration until the uncertainty of prediction point is lower than the preset threshold or the total sampling point number reaches the preset percentage of sampling unit point number, finally complete echo data sparse acquisition and sparse sampling data matrix construction.The method of the application constructs control matrix based on target structure information, while retaining target scattering information to the greatest extent and eliminating redundant sampling points, effectively improves the imaging quality.

Description

Technical Field

[0001] This invention belongs to the field of microwave imaging technology, focusing on the rapid acquisition of the scattering signal of the measured target in near-field three-dimensional imaging, specifically involving a sparse acquisition method for echo data in microwave near-field three-dimensional imaging. Background Technology

[0002] One method for achieving near-field 3D microwave imaging is planar scanning imaging. This involves mounting microwave transceiver antennas on a scanning rig and controlling the rig to move at equal intervals in the horizontal and vertical directions according to the Nyquist sampling theorem, forming a virtual planar scanning spatial sampling grid. Backscattering data of the target object, including amplitude and phase, is recorded on each spatial sampling grid. The final measurement results are recorded in the form of a 3D matrix, with each matrix element corresponding to the single-frequency echo signal data of a spatial sampling point. In this system, the resolution of the acquired 3D image along the horizontal and vertical directions of the scanning plane depends on the size of the scanning plane; that is, the larger the scanning plane, the higher the resolution. However, within the framework of the Nyquist sampling theorem, a large-scale scanning plane leads to low data acquisition efficiency and causes the acquired sampling data to grow quadratically, resulting in a massive amount of acquired echo data.

[0003] One effective way to reduce the data sampling rate is sparse sampling, which involves sparse data acquisition in a planar scanning space and using sparse signal recovery theory to achieve three-dimensional imaging. This theory states that sparse downsampling of a sparsely reflected signal (or a signal that is sparse in a certain transform domain, such as the time domain, spatial domain, frequency domain, or polarization domain) yields a small amount of sampled data containing a large amount of target information. Sparse signal recovery techniques are then used to effectively approximate and recover the original signal. By extracting target feature information, the scattering characteristics, shape contours, spatial location, and other physical and geometric features of the target can be obtained. Microwave near-field three-dimensional sparse imaging relies on the sparse characteristics of the target represented by microwave signals, overcoming the constraints of the traditional Nyquist sampling theorem on the amount of data acquired. By designing an incoherent sparse sampling matrix, microwave reflected signals are acquired with a number of sampling points far lower than required by the Nyquist sampling theorem. Then, a nonlinear optimization algorithm is used to achieve high-precision reconstruction of the microwave image of the target from the small amount of sampled data. This technique can significantly reduce the amount of data acquired and the system complexity while ensuring imaging quality.

[0004] In sparse imaging, the quality of scattering data acquired through sparse sampling is crucial for imaging. Current sparse sampling data acquisition is achieved through sparse sampling control matrices, which project high-dimensional microwave scattering signals onto low-dimensional measurement vectors by constructing a suitable sparse sampling control matrix. However, sparse imaging theory requires that the designed sparse sampling control matrix possess good non-correlation properties, meaning that the sampling points in the sparse space exhibit good randomness. Currently, sparse sampling control matrices are mainly divided into random sparse sampling matrices and deterministic sparse sampling matrices. Many sparse sampling control matrices have been proven to meet the non-correlation requirement and can effectively recover signals, but they still have the following problems: 1) Random sparse sampling matrices are effective in simulations but difficult to implement in engineering, resulting in significant uncertainty; 2) Traditional sampling matrix construction methods, due to a lack of consideration of the target's scattering characteristics, are prone to losing key target information, thus affecting the final imaging quality. Summary of the Invention

[0005] To overcome the limitations of existing sparse sampling methods in target scattering data acquisition, such as difficulties in engineering implementation, loss of key information, and low sampling efficiency, this invention proposes a sparse acquisition method for echo data in microwave near-field three-dimensional imaging. This method estimates the amplitude and uncertainty of the reflected signal from the target, thereby selecting the scanning area with the highest correlation and selecting spatial sparse sampling points for near-field three-dimensional sparse imaging in real time. Compared to traditional sampling matrix construction methods, this method combines the scattering characteristics of the target and selects the required sampling points based on these characteristics. It maximizes the preservation of target scattering information while eliminating redundant sampling points, thereby improving imaging quality, reducing the number of samples required for image reconstruction, and shortening data acquisition time.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A sparse acquisition method for echo data in microwave near-field three-dimensional imaging includes the following steps:

[0008] Step 1: Planar Scanning Region Segmentation: First, the scanning plane of the planar scanning imaging is discretized in two dimensions. Let the planar scanning region formed by the antenna movement be... , and These are the scanning ranges in two dimensions of the scanning plane (i.e., the horizontal and vertical directions); the resolution of the imaging system in the horizontal and vertical directions is determined by the scanning range. and The wavelength λ corresponding to the continuous wave frequency and the vertical distance from the scanning plane to the target being measured. The horizontal resolution of the imaging system is jointly determined; thus, it is determined. and vertical resolution They are respectively:

[0009]

[0010]

[0011] , The selection of the sampling interval must meet the resolution requirements of the imaging system; furthermore, to satisfy the sampling theorem, the sampling interval must be... and The following inequalities must be satisfied:

[0012]

[0013]

[0014] in, and They represent the measured target along... x shaft and y Maximum dimension of the axis; according to sampling intervals in the horizontal and vertical directions respectively. and Equal-interval sampling will yield... The sampling point units are non-overlapping; then the entire planar scanning area is divided into sub-regions, with the horizontal direction equally divided into... part( Less than ), divided vertically into equal parts part( Less than ), divided into Each subregion contains several sampling point units; parameters and The selection of sampling points should follow the following requirements: the total number of sampling points in each sub-region should be strictly controlled within a reasonable range of 50-300, while also taking into account the balance between prediction accuracy and computational efficiency.

[0015] Step 2: Initial Sampling Point Selection and Data Acquisition: Based on the correspondence between the spatial location of the target and the scanning area, select a set of sampling points of size [missing information] from each sub-region. (The size of this matrix is ​​equal to the total number of sampling points in the corresponding sub-region) The initial sampling points form the initial sampling point coordinate matrix. ,in, Represents a matrix. The dimension is Row 2, Column 2 This indicates the spatial sampling position coordinates of the antenna in the scanning plane. , This yields the sampled data matrix. , The dimension is Row 1, Column 1, containing sampled data Expressed as follows:

[0016]

[0017] In the formula, x , y, z Indicates the coordinates of the scattering point; Represents a rectangular function; Indicates An exponential function with base 0; This represents the distance between the transmitting and receiving antennas and the scattering point; Indicates that it is located at The scattering coefficient at the scattering point; c At the speed of light, For time; For the pulse center at Frequency of time, The center frequency of the pulse. , This indicates the frequency step interval in each pulse. T Indicates the pulse duration; H Indicates the number of pulses in the pulse train;

[0018] Step 3: Intelligent selection of sparse sampling points based on the structural information of the target being measured:

[0019] Step 3-1: Observation Model Setting: Construct an independent observation model for each sub-region; for the initial sampling point coordinates in the sub-region, i.e., input... , , ..., The corresponding sampled data in the sub-region is the output. Assume:

[0020]

[0021] The above formula is the observation model, where, It is independent and identically distributed Gaussian noise. This indicates that the mean is 0 and the variance is . Gaussian distribution; The true function values ​​given by the Gaussian process follow the following Gaussian distribution:

[0022] ,

[0023] in, express obey The distribution, Represents a Gaussian process. The coordinates of another initial sampling point in the sub-region. It is a mean function; Let covariance be the function, and its expression is as follows:

[0024]

[0025] in, Represents the square of the norm. For signal variance, For length scale;

[0026] Step 3-2: Construct a joint Gaussian distribution using the obtained initial sampling points. and corresponding sampling data The new input point, i.e., the prediction point, is predicted using the following method. Corresponding predictive sampling data ; Assumption and The joint distribution of is a multivariate Gaussian distribution, with the following form:

[0027]

[0028] in, It is the initial sampling point The covariance matrix between them, the covariance matrix The elements are determined by the covariance function. Given:

[0029]

[0030] I It is the identity matrix. It is the initial sampling point and prediction points The covariance matrix between them , represents a matrix transpose, It is a prediction point The autocovariance;

[0031] Step 3-3: Calculate the conditional mean and variance of the joint Gaussian distribution, and obtain the conditional mean of the predicted distribution according to the conditional distribution formula of the Gaussian distribution. and variance as follows:

[0032]

[0033]

[0034] right By performing the square root operation, we can obtain the new input point, i.e., the prediction point. Corresponding prediction sampling data uncertainty ;

[0035] Steps 3-4: Uncertainty-guided adaptive iterative sampling mechanism: Uncertainty is sampled separately in each sub-region. Sort and select (Experience shows that when the total number of samples in a sub-region is less than 100, Prioritize taking 1-4 small batches; if the total number of samples in a sub-region is greater than or equal to 100. Prioritize selecting 5-10 sampling points with the highest uncertainty (in a large batch); all selected sampling points in all sub-regions are used as sparse sampling points to form a coordinate matrix. ;in, The dimension is Row 2, Column 2 This indicates the spatial sampling position coordinates of the antenna in the scanning plane. ;

[0036] Step 4: Scan path planning based on genetic algorithm

[0037] Step 4-1: Using path encoding (Permutation Encoding), the sparse sampling points selected in Step 3 are numbered from 1 to 1 using decimal encoding. ;

[0038] Step 4-2: Generate a file containing... A population of individuals, with a population size of _____. Each individual in the population represents a sparse sampling point scanning path; the fitness function value is calculated, which is the reciprocal of the total distance of the sparse sampling point scanning paths corresponding to each individual in the population, with maximizing the fitness function value as the optimization objective; the distance between adjacent numbered sampling positions is calculated as follows:

[0039]

[0040] The fitness function value is:

[0041]

[0042] in, For the first The fitness of an individual This indicates the distance between the first adjacent numbered sampling positions. To the Distance between adjacent numbered sampling positions The sum of the sums, ;

[0043] Step 4-3: The selection operation uses Roulette Wheel Selection to calculate the sum of the fitness of all individuals in the population.

[0044]

[0045] Calculate the selection probability for each individual. Generate a random number between 0 and 1. Starting with the first individual, the selection probability of each individual is accumulated sequentially. When the accumulated sum exceeds... When the population reaches its maximum size, the last individual in the cumulative count is selected to enter the next generation of the population; this process is repeated until the number of individuals selected is equal to the population size.

[0046] Step 4-4: The crossover operation uses the Partially Mapped Crossover (PMX) method. Two sampling points are randomly selected from the scan path corresponding to parent individual 1, and the scan path segment between these two sampling points is retained in child individual 1. Following the scan path corresponding to parent individual 2, the unfilled scan path segments in child individual 1 are sequentially filled in. If a duplicate scan path segment is encountered, a mapping relationship is established to replace the duplicate scan path segment with a corresponding unused scan path segment in parent individual 1, until the scan path of child individual 1 is complete. The same operation is performed on parent individual 2 to generate child individual 2.

[0047] Steps 4-5: The mutation operation uses a multi-point mutation method to ensure that the randomly searched solutions cover the entire space; two sampling points are randomly selected in the scan path corresponding to the parent individual, and their positions in the scan path are swapped to generate new offspring individuals;

[0048] Steps 4-6: Termination condition: Set the maximum number of evolutions. When the algorithm reaches the maximum number of evolutions, stop running. At this time, the scan path corresponding to the individual with the largest fitness function value is the planned scan path.

[0049] Following the planned scanning path, scan the sparse sampling points in step 3 to obtain the sparse sampling data matrix. This is incorporated into the sampling data matrix in step 2. A new sparse sampling data matrix is ​​obtained. , ,in Indicates matrix concatenation;

[0050] Step 5: Iteration Termination Determination: Repeat steps 2 to 4 until the maximum uncertainty is reached. When the number of samples falls below a preset threshold or the total number of sampling points m reaches 50% of the total number of sampling point units, the sparse sampling point data acquisition and sparse sampling data matrix construction are completed.

[0051] Beneficial effects

[0052] This invention proposes a sparse acquisition method for echo data in microwave near-field 3D imaging. The method first divides the entire scanning plane into several sub-regions and processes them in parallel, reducing the overall complexity of the observation model and significantly decreasing computational load, laying the foundation for efficient acquisition. Within each sub-region, an observation model is constructed. Using initial sampling points and their corresponding sampling data, the method accurately predicts the sampling data and prediction uncertainty at other unknown locations within the sub-region, prioritizing the acquisition of data from areas with higher prediction uncertainty. This constructs an adaptive sampling mechanism that incorporates the scattering characteristics of the target, effectively avoiding the loss of key scattering information. After determining the next sampling points, a genetic algorithm is used to plan the sampling path, significantly reducing the time consumption associated with traditional row-by-row scanning and minimizing the uncertainty caused by background changes during long-term scanning, further improving the reliability of the sampling data. After acquiring new sampling data, the observation model is updated in real-time, achieving a refined construction of the sparse sampling data matrix, ultimately ensuring superior quality in microwave near-field 3D imaging. Attached Figure Description

[0053] Figure 1 This is a flowchart of the algorithm of the present invention.

[0054] Figure 2 The target being measured.

[0055] Figure 3 This is a segmentation map of the scanned region.

[0056] Figure 4 The initial sampling point selection diagram.

[0057] Figure 5 This is a map showing the predicted sampling points.

[0058] Figure 6 The scan path diagram planned for the genetic algorithm.

[0059] Figure 7 is the sparse sampling control matrix constructed.

[0060] Figure 8 Autocorrelation analysis of the sparse sampling control matrix.

[0061] Figure 9 This is an image showing the imaging effect. Detailed Implementation

[0062] The present invention will be further described in detail below with reference to embodiments and accompanying drawings:

[0063] by Figure 1 Based on the flowchart of the method of this invention, Figure 2 The structural features of the target object are shown, and the method process of the present invention is described in detail.

[0064] Step 1: Planar Scanning Region Segmentation. First, the scanning plane for planar scanning imaging is discretized in two dimensions. The X-band is used, with frequencies incremented in 40MHz steps. The scanning range in both the horizontal and vertical directions is 1m. The planar scanning region formed by antenna movement is... The center frequency of the X-band electromagnetic wave corresponds to a wavelength of 0.03m, and the vertical distance from the scanning plane to the target is 2m. Therefore, the horizontal and vertical resolutions are determined as follows: , .

[0065] Furthermore, to satisfy the sampling theorem, the sampling interval... and Must meet , By sampling at equal intervals of 0.02m in both the horizontal and vertical directions, 51*51 non-overlapping sampling point units can be obtained. For example... Figure 3 As shown, the two-dimensional region is divided into sub-regions, with the horizontal direction divided into 3 equal segments and the vertical direction divided into 3 equal segments, resulting in 9 sub-regions, each containing 289 sampling points.

[0066] Step 2: Initial Sampling Point Selection and Data Acquisition: Based on the correspondence between the spatial location of the target and the scanning area, such as... Figure 4 As shown, five initial sampling points are selected from each sub-region to form the initial sampling point coordinate matrix. ,in, Represents a matrix. The dimensions are 5 rows and 2 columns. This indicates the spatial sampling position coordinates of the antenna in the scanning plane. , This yields the sampled data matrix. , The dimension is 5 rows and 1 column, where the sampled data Expressed as follows:

[0067]

[0068] In the formula, x , y, z Indicates the coordinates of the scattering point; Represents a rectangular function; Indicates An exponential function with base 0; This represents the distance between the transmitting and receiving antennas and the scattering point; Indicates that it is located at The scattering coefficient at the scattering point; c At the speed of light, For time; For the pulse center at Frequency of time, The center frequency of the pulse. , This indicates the frequency step interval in each pulse. T Indicates the pulse duration; H This indicates the number of pulses in the pulse train.

[0069] Step 3: Intelligent selection of sparse sampling points based on the structural information of the target being measured:

[0070] Step 3-1: Observation Model Setting: Construct an independent observation model for each sub-region; for the initial sampling point coordinates in the sub-region, i.e., input... , , ..., The corresponding sampled data in the sub-region is the output. Assume:

[0071]

[0072] The above formula is the observation model, where, It is independent and identically distributed Gaussian noise. This indicates that the mean is 0 and the variance is . Gaussian distribution; The true function values ​​given by the Gaussian process follow the following Gaussian distribution:

[0073] ,

[0074] in, express obey The distribution, Represents a Gaussian process. The coordinates of another initial sampling point in the sub-region. It is a mean function; Let covariance be the function, and its expression is as follows:

[0075]

[0076] in, Represents the square of the norm. For signal variance, For length scale;

[0077] Step 3-2: Construct a joint Gaussian distribution using the obtained initial sampling points. and corresponding sampling data The new input point, i.e., the prediction point, is predicted using the following method. Corresponding predictive sampling data ; Assumption and The joint distribution of is a multivariate Gaussian distribution, with the following form:

[0078]

[0079] in, It is the initial sampling point The covariance matrix between them, the covariance matrix The elements are determined by the covariance function. Given:

[0080]

[0081] I It is the identity matrix. It is the initial sampling point and prediction points The covariance matrix between them , represents a matrix transpose, It is a prediction point The autocovariance;

[0082] Step 3-3: Calculate the conditional mean and variance of the joint Gaussian distribution, and obtain the conditional mean of the predicted distribution according to the conditional distribution formula of the Gaussian distribution. and variance as follows:

[0083]

[0084]

[0085] right By performing the square root operation, we can obtain the new input point, i.e., the prediction point. Corresponding prediction sampling data uncertainty ;

[0086] Steps 3-4: Uncertainty-guided adaptive iterative sampling mechanism: such as Figure 5 As shown, the uncertainty is considered in each sub-region. The sampling points are sorted, and the five with the highest uncertainty are selected. All selected sampling points in all sub-regions are then used as sparse sampling points to form a coordinate matrix. ;in, The dimension is 45 rows and 2 columns. This indicates the spatial sampling position coordinates of the antenna in the scanning plane. .

[0087] Step 4: Scan path planning based on genetic algorithm

[0088] Step 4-1: Use path encoding (Permutation Encoding) to number the sparse sampling points selected in Step 3 from 1 to 45 using decimal encoding;

[0089] Step 4-2: Generate a population of 450 individuals, where each individual represents a sparse sampling point scan path; calculate the fitness function value, which is the reciprocal of the total distance of the sparse sampling point scan paths corresponding to each individual in the population, with maximizing the fitness function value as the optimization objective; the distance between adjacent numbered sampling positions is calculated as follows:

[0090]

[0091] The fitness function value is:

[0092]

[0093] in, For the first The fitness of an individual This indicates the distance between the first adjacent numbered sampling positions. Distance to the 44th adjacent sampling position The sum of the sums, ;

[0094] Step 4-3: The selection operation uses Roulette Wheel Selection to calculate the sum of the fitness of all individuals in the population.

[0095]

[0096] Calculate the selection probability for each individual. Generate a random number between 0 and 1. Starting with the first individual, the selection probability of each individual is accumulated sequentially. When the accumulated sum exceeds... When the population reaches its maximum size, the last individual in the cumulative count is selected to enter the next generation of the population; this process is repeated until the number of individuals selected is equal to the population size.

[0097] Step 4-4: The crossover operation uses the Partially Mapped Crossover (PMX) method. Crossover is performed on individuals with a probability of 0.6. Two sampling points are randomly selected from the scan path corresponding to parent individual 1, and the scan path segment between these two sampling points is retained in child individual 1. Following the scan path corresponding to parent individual 2, the unfilled scan path segments in child individual 1 are sequentially filled in. If a duplicate scan path segment is encountered, a mapping relationship is established to replace the duplicate scan path segment with a corresponding unused scan path segment in parent individual 1, until the scan path of child individual 1 is complete. The same operation is performed on parent individual 2 to generate child individual 2.

[0098] Steps 4-5: The mutation operation uses a multi-point mutation method to randomly search the entire space where solutions may exist; the mutation operation is performed with a probability of 0.2, randomly selecting two sampling points in the scan path corresponding to the parent individual, and swapping their positions in the scan path to generate a new offspring individual;

[0099] Steps 4-6: Termination condition. Set the maximum number of evolutions to 1000. When the algorithm reaches the maximum number of iterations, stop running. Figure 6 As shown, the scanning path corresponding to the individual with the highest fitness is the planned scanning path; scanning sampling points according to the planned path saves a lot of time compared to the traditional row-by-row and column-by-column scanning.

[0100] Following the planned scanning path, scan the sparse sampling points in step 3 to obtain the sparse sampling data matrix. This is incorporated into the sampling data matrix in step 2. A new sparse sampling data matrix is ​​obtained. , ,in Indicates matrix concatenation;

[0101] Step 5: Iteration Termination Determination: Repeat steps 2 to 4 until the maximum uncertainty is reached. Less than 8 or the total number of sampling points reaches the total number of sampling point units. ,like Figure 7 As shown, the uncertainty of all sub-region prediction sampling points is less than 8, at which point the sparse sampling point data acquisition and sparse sampling data matrix construction are completed.

[0102] The sparse sampling data matrix used for sparse 3D imaging needs to meet the requirement of non-correlation, such as Figure 8 As shown, the autocorrelation of the constructed sparse sampling data matrix is ​​calculated, and the autocorrelation satisfies the requirements of sparse imaging.

[0103] Sparse imaging is performed using a constructed sparse sampling data matrix, and the imaging effect is as follows: Figure 9 As shown, the image quality is good and can reflect... Figure 2 The spatial location and structure of the target under test demonstrate the effectiveness of the method of the present invention.

Claims

1. A method for sparse acquisition of echo data for microwave near-field three-dimensional imaging, characterized in that: Includes the following steps: Step 1: planar scanning region segmentation: the scanning plane of planar scanning imaging is discretized in two dimensions, and is sampled at equal intervals in the horizontal direction and the vertical direction respectively to obtain a plurality of non-overlapping sampling point units; then the entire planar scanning region is divided into sub-regions, and each sub-region contains a plurality of sampling point units; and ​​ Step 2: Initial sampling point selection and sampling data acquisition to obtain the sampling data matrix; Step 3: Intelligent selection of sparse sampling points based on the structural information of the target being measured: Construct independent observation models for each sub-region; where, is the initial sampling point coordinate in the sub-region, i.e., the input, is the corresponding sampling data in the sub-region, i.e., the output, is an independent and identically distributed Gaussian noise, represents a Gaussian distribution with a mean of 0 and a variance of ; and is a true function value given by the Gaussian process, subject to a Gaussian distribution as follows: wherein denotes subject to a distribution, denotes a Gaussian process, is the coordinate of another initial sampling point in the sub-region, is the mean function; is the covariance function; Construct a joint Gaussian distribution, using the obtained initial sampling points and corresponding sampling data, predict the prediction sampling data corresponding to the new input point, i.e., the prediction point; obtain the conditional mean of the prediction distribution according to the conditional distribution formula of the Gaussian distribution And variance , square root operation is performed on the variance , to obtain the uncertainty of the prediction sampling data corresponding to the new input point, i.e., the prediction point ; In each sub-region, the uncertainty of each sampling point is ranked, and the sampling point with the largest uncertainty is selected. The selected sampling points in all sub-regions are used as sparse sampling points to form a coordinate matrix.​ Step 4: Genetic algorithm-based scan path planning: adopt path encoding mode, number the sparse sampling points selected in step 3 in the form of decimal encoding, generate a population containing individuals, each individual in the population represents a sparse sampling point scan path; calculate the fitness function value, with the maximum fitness function value as the optimization goal; the selection operation adopts the roulette selection method; the crossover operation adopts the partial mapping crossover method; the mutation operation adopts the multi-point mutation mode; set the maximum evolution number, when the algorithm iteration number reaches the maximum evolution number, stop running; at this time, the scan path corresponding to the individual with the maximum fitness function value is the planned scan path; scan the sparse sampling points in step 3 according to the planned scan path, obtain the sparse sampling data matrix, and merge it into the sampling data matrix in step 2 to obtain a new sparse sampling data matrix; Step 5: Iteration termination decision: repeat steps 2 to 4 until the maximum uncertainty Below the preset threshold or the total number of sampling points m Reaching the preset percentage of the total number of sampling point units M 0, at which time the sparse sampling point data acquisition and sparse sampling data matrix construction are completed.

2. The method for sparse acquisition of echo data for microwave near-field three-dimensional imaging according to claim 1, characterized in that: The step 1 is specifically as follows: firstly, the scanning plane of the planar scanning imaging is two-dimensionally discretized, and the planar scanning region formed by the antenna movement is , and are the scanning ranges of the two dimensions of the scanning plane, i.e. the horizontal direction and the vertical direction; the resolutions of the imaging system in the horizontal direction and the vertical direction are determined by the scanning ranges and , the wavelength λ corresponding to the continuous wave frequency, and the vertical distance from the scanning plane to the measured target; the resolutions of the imaging system in the horizontal direction and the vertical direction determined in this way are respectively: and . , The selection of the sampling interval must satisfy the resolution requirement of the imaging system; in addition, to satisfy the sampling theorem, the sampling interval and must satisfy the following inequality: wherein, and respectively represent the maximum dimension of the measured target along the x axis and the y axis; in the horizontal direction and the vertical direction, respectively, equal-interval sampling is performed at sampling intervals and , obtaining non-overlapping sampling point units; then, sub-region division is performed on the entire planar scanning region, the horizontal direction is equally divided into segments, smaller than , and the vertical direction is equally divided into segments, smaller than , and the division obtains sub-regions, each of which contains a plurality of sampling point units; the selection of parameters and needs to follow the following requirements: the total number of sampling points in each sub-region needs to be controlled within the interval of 50-300, while the balance between prediction accuracy and calculation efficiency needs to be considered.

3. The method for sparse acquisition of echo data for microwave near-field three-dimensional imaging according to claim 1, characterized in that: Step 2 is specifically as follows: according to the corresponding relationship between the spatial position of the measured target and the scanning region, a group of initial sampling points with a size of is selected from each sub-region to form an initial sampling point coordinate matrix , wherein represents a matrix, the dimension of the matrix is 2 rows and 2 columns, represents the spatial sampling position coordinates of the antenna in the scanning plane, , and thus a sampling data matrix is obtained, the dimension of the matrix is 1 row and 1 column, wherein the sampling data is expressed by the following formula: wherein x , y, z denotes the scattering point position coordinate; denotes the rectangular function; denotes the exponential function with base ; denotes the distance between the transceiver antenna and the scattering point; denotes the scattering coefficient of the scattering point located at ; c is the speed of light, is the time; is the frequency of the pulse center at the time , is the center frequency of the pulse, , denotes the frequency step interval in each pulse, T denotes the pulse duration; H denotes the number of pulses in the pulse train.

4. The echo data sparsely sampled method for three-dimensional imaging in the near field of microwaves according to claim 1, characterized in that: Step 3 specifically includes the following steps: Step 3-1: Observation model setting: build independent observation model in each sub-region respectively; for the initial sampling point coordinates in the sub-region, i.e. input , ,… and the corresponding sampling data in the sub-region, i.e. output , hypothesis: The above equation is the observation model, where, is an independent and identically distributed Gaussian noise, represents a Gaussian distribution with mean 0 and variance ; and is the true function value given by a Gaussian process, which follows a Gaussian distribution as follows: ， wherein denotes subject to a distribution, denotes a Gaussian process, is the coordinate of another initial sampling point in the sub-region, is the mean function; is the covariance function, which is expressed as follows: wherein denotes the square of the norm, is the signal variance, is the length scale; Step 3-2: Construct a joint Gaussian distribution using the obtained initial sampling points. and corresponding sampling data The new input point, i.e., the prediction point, is predicted using the following method. Corresponding predictive sampling data ; Assumption and The joint distribution of is a multivariate Gaussian distribution, with the following form: where is the initial sample point is the covariance matrix between the initial sample point and the sample point is given by the covariance function I is the identity matrix, is the initial sampling point and the prediction point is the covariance matrix between the initial sampling point , denotes the transpose of the matrix is the prediction point is the auto-covariance of the prediction point . Step 3-3: Compute the conditional mean and variance of the joint Gaussian distribution, the conditional mean of the predictive distribution is obtained according to the conditional distribution formula of the Gaussian distribution and variance as follows: Taking square root operation, the new input point, i.e. the prediction point corresponding to the prediction sampling data of the uncertainty ;​ Steps 3-4: Uncertainty-guided adaptive iterative sampling mechanism: Uncertainty is sampled separately in each sub-region. Sort and select The sampling point with the highest uncertainty; all selected sampling points in all sub-regions are used as sparse sampling points to form a coordinate matrix. ;in, The dimension is Row 2, Column 2 This indicates the spatial sampling position coordinates of the antenna in the scanning plane. .

5. The method for sparse acquisition of echo data for microwave near-field three-dimensional imaging according to claim 4, characterized in that: When the total number of samples in a sub-region is less than 100 Select 1-4; if the total number of samples in the sub-region is greater than or equal to 100, Take 5-10.

6. The method for sparse acquisition of echo data for microwave near-field three-dimensional imaging according to claim 1, characterized in that: In step 4, the fitness function value is the reciprocal of the total distance of the scan path of the sparse sampling points corresponding to each individual in the population.

7. The method for sparse acquisition of echo data for microwave near-field three-dimensional imaging according to claim 1, characterized in that: In step 4, the partial mapping cross method is as follows: two sampling points are randomly selected in the scanning path corresponding to parent individual 1, and the scanning path segment between the two sampling points is retained in child individual 1; according to the scanning path corresponding to parent individual 2, the unfilled scanning path segments in child individual 1 are filled in sequentially; if a case of duplicate scanning path segments is encountered, a mapping relationship is established to replace the duplicate scanning path segments with the corresponding unused scanning path segments in parent individual 1, until the scanning path of child individual 1 is complete; the same operation is performed on parent individual 2 to generate child individual 2.

8. The method for sparse acquisition of echo data for microwave near-field three-dimensional imaging according to claim 1, characterized in that: In step 5, the preset percentage is used. M 0 represents 50%.

Images