Hidden target three-dimensional imaging method and system

Abstract

The invention provides a hidden target three-dimensional imaging method and system, and the method comprises the steps: S1, transmitting an electromagnetic wave signal, and receiving an echo signal of a target; the target is a hidden target; s2, imaging the azimuth-distance-height of the whole detection scene based on the echo signal, and reflecting the position, contour and state information of the target; s3, locking the position of the target in the azimuth-distance two-dimensional plane, and selecting a range region where the target is located as a three-dimensional imaging region; and S4, obtaining a target three-dimensional imaging result according to imaging of the three-dimensional imaging area. According to the three-dimensional fine imaging method provided by the invention, the two-dimensional imaging result is firstly detected and positioned, and then the three-dimensional imaging is used as the rear-end network input, so that compared with a direct three-dimensional imaging method, the model parameter quantity is reduced, and the hardware processing efficiency, the algorithm processing speed and the recognition precision are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional imaging and target recognition of wall-penetrating radar, and in particular, relates to a method and system for three-dimensional imaging of concealed targets, and more particularly, to a method for three-dimensional fine imaging of concealed targets. Background Art

[0002] With the continuous development of three-dimensional imaging technology, the three-dimensional information of distance-azimuth-altitude has been widely pursued by researchers when detecting concealed targets. Therefore, imaging methods such as synthetic aperture imaging radar technology, inverse scattering, time reversal and back projection have gradually emerged.

[0003] Time-frequency analysis technology is one of the most commonly used tools for penetrating-wall radar to analyze the posture characteristics of hidden targets inside walls. This method extracts the target echo characteristics caused by the target posture by projecting the echo energy onto the two-dimensional time-frequency plane and forms a characteristic representation. However, when electromagnetic waves penetrate walls, reflection, refraction, velocity change, and energy attenuation will occur. Traditional methods are difficult to solve problems such as offset, defocusing, or false targets in the imaging results.

[0004] After the rise of deep learning technology, deep convolutional neural networks have begun to be applied to the feature recognition of three-dimensional imaging targets. However, the application research of existing deep learning methods in this field considers more the design issues of network structure and training and learning methods. In order to improve the spatial imaging resolution of the target under limited antenna aperture, some scholars have applied convolution super-resolution technology to three-dimensional imaging, but this method has high requirements for the echo signal-to-noise ratio and is currently difficult to apply to complex environments. Subsequently, some people proposed an integrated feature recognition algorithm and conducted experimental verification in a variety of complex environments. Some scholars have used deconvolution, deep neural networks and other methods to achieve three-dimensional image enhancement for grating sidelobe suppression and mainlobe sharpening of radar images. However, existing deep learning models have high data set dependence, high computational complexity, and cannot effectively deal with problems such as scarce training data and limited computing resources.

[0005] At present, the existing deep neural networks have problems such as large number of model parameters, high computational complexity, large data requirements, and low spatial imaging resolution when imaging hidden targets.

[0006] Patent document CN113050083A discloses a method for reconstructing human body posture based on point cloud by ultra-wideband radar, which belongs to the field of radar signal processing. The present invention aims at the problem of poor readability of ultra-wideband radar imaging results. By first receiving the radar echo signal, the detection scene is imaged in two dimensions in the range-azimuth direction; the two-dimensional imaging result is preprocessed to lock the target position; then the target area is imaged in three dimensions, the scattering intensity isosurface is constructed, and the feature points are sampled to obtain the point cloud data of the human target; the point cloud data is input into the trained neural network to reconstruct the posture of the human target; finally, the detected target position is fused with the posture reconstruction result to obtain the posture of the target at the real position. The present invention combines the radar signal processing method with the deep learning method, and uses a lightweight neural network to quickly reconstruct the posture of the human target from the radar point cloud data, providing system users with richer human target detail information. This solution cannot solve the problems of large model parameters, high computational complexity, and low spatial imaging resolution when imaging hidden targets. Summary of the invention

[0007] In view of the defects in the prior art, an object of the present invention is to provide a method and system for three-dimensional imaging of a concealed target.

[0008] A method for three-dimensional imaging of a concealed target provided by the present invention comprises:

[0009] Step S1: transmitting an electromagnetic wave signal and receiving an echo signal of the target;

[0010] Step S2: Based on the echo signal, the azimuth-distance-height of the entire detection scene is imaged to reflect the position, outline and status information of the target;

[0011] Step S3: locking the position of the target on the azimuth-distance two-dimensional plane, and selecting the range area where the target is located as the three-dimensional imaging area;

[0012] Step S4: imaging the three-dimensional imaging area to obtain a target three-dimensional imaging result.

[0013] Preferably, in step S1: transmitting an electromagnetic wave signal by an ultra-wideband radar;

[0014] In the step S1, suppressing the wall echo in the echo signal;

[0015] In the step S2, static clutter of the echo signal is filtered out;

[0016] In step S2, the azimuth-distance-height of the entire detection scene is imaged by a two-dimensional BP imaging algorithm, namely, a BP algorithm, and the steps include:

[0017] Step A1: performing distance matching on the echo signal by using the BP algorithm, and then performing inverse Fourier transform on the echo signal to obtain a change result;

[0018] Step A2: Obtain the time delay t of the transmitting and receiving antenna combination of the ultra-wideband radar, and obtain the target function based on the change result; and form an azimuth-range-altitude image of the detection scene according to the target function.

[0019] Preferably, in step S3, the probability of the target appearing is detected; the judgment condition of the detection result is expressed mathematically as follows:

[0020]

[0021] Where x(t) represents the input signal; n(t) represents the system noise; s(t) represents the received signal; t represents time t;

[0022] The probability of false alarm during detection is expressed mathematically as:

[0023]

[0024] Among them, P(D 1 |H 0 ) represents the probability of false alarm; D 1 Indicates making a detection decision; H 0 Indicates no signal; V T represents the decision threshold; p(x|H 0 ) means that under the assumption that H 0 The probability of making a detection decision x when is true; x represents the decision that the system can make;

[0025] Among them, V T According to the assumption P(D 1 |H 0 )=P fa Decision; P fa Indicates the false alarm rate; when the judgment result is H 1 When , the mathematical expression of probability is:

[0026]

[0027] Among them, H 1 Indicates that there is a signal;

[0028] The judgment rules are:

[0029]

[0030] Among them, Λ 0 Indicates the set threshold value; H 1 (Λ 0 ∝VT ) indicates the case with a target; ∝ indicates the sign of being proportional to .

[0031] Preferably, the step S4 includes:

[0032] Step S4.1: gridding the three-dimensional imaging area, and treating each grid as a scattering point;

[0033] Step S4.2: Calculate the distance from the scattering point to the transceiver antenna of the ultra-wideband radar; and calculate the time delay between the transceiver signal and the receiver signal according to the distance;

[0034] Step S4.3: obtaining data of corresponding pixel points in the one-dimensional range image of the channel echo corresponding to the time delay;

[0035] Step S4.4: traverse all the transceiver channels, accumulate the addressing results to obtain the pixel value of the grid point, and then traverse each of the grid points to obtain the final offset imaging result, that is, the target three-dimensional imaging result;

[0036] In step S4.4, the target three-dimensional imaging result is expressed as follows:

[0037]

[0038] Where I(x, y, z) represents the target three-dimensional imaging result; R jk Indicates the antenna measurement point (x j ,y k ,0) to the imaging point (x,y,z); J represents the maximum value of x movement; K represents the maximum value of y movement; v represents the propagation speed of electromagnetic waves in underground media; Represents the round-trip delay of electromagnetic waves from the imaging point to the corresponding antenna position;

[0039] The R jk The mathematical expression is:

[0040]

[0041] Among them, x, y and z represent the three-dimensional coordinates of the target three-dimensional imaging; j With y k Respectively represent the abscissa and ordinate of the antenna measurement point;

[0042] The calculation amount of the target three-dimensional imaging result is expressed as follows:

[0043] T BP =N g (N a T td +(N a -1)Tadd )

[0044] Among them, T BP Indicates the amount of calculation; N g Indicates the number of grid points; N a Indicates the number of time delay solutions; T td Indicates the time to solve a delay; T add Indicates the time to perform one addition operation.

[0045] Preferably, in step S4, artifacts of the target three-dimensional imaging result are reduced by a convolutional neural network, and a target recognition result is output.

[0046] A concealed target three-dimensional imaging system provided by the present invention comprises:

[0047] Module M1: emits electromagnetic wave signals and receives echo signals from the target;

[0048] Module M2: Based on the echo signal, the azimuth-distance-height of the entire detection scene is imaged to reflect the position, outline and status information of the target;

[0049] Module M3: lock the position of the target in the azimuth-distance two-dimensional plane, and select the range area where the target is located as the three-dimensional imaging area;

[0050] Module M4: obtains a target three-dimensional imaging result according to the three-dimensional imaging area imaging.

[0051] Preferably, in the module M1: an electromagnetic wave signal is emitted by an ultra-wideband radar;

[0052] In the module M1, suppressing the wall echo in the echo signal;

[0053] In the module M2, static clutter of the echo signal is filtered out;

[0054] In the module M2, the azimuth-distance-height of the entire detection scene is imaged by a two-dimensional BP imaging algorithm, namely, a BP algorithm, and the steps include:

[0055] Module A1: performing distance matching on the echo signal by using the BP algorithm, and then performing inverse Fourier transform on the echo signal to obtain a change result;

[0056] Module A2: Obtain the time delay t of the transmitting and receiving antenna combination of the ultra-wideband radar, and obtain the target function based on the change result; and form an azimuth-range-altitude image of the detection scene according to the target function.

[0057] Preferably, in the module M3, the probability of the target appearing is detected; the judgment condition of the detection result is expressed as follows:

[0058]

[0059] Where x(t) represents the input signal; n(t) represents the system noise; s(t) represents the received signal; t represents time t;

[0060] The probability of false alarm during detection is expressed mathematically as:

[0061]

[0062] Among them, P(D 1 |H 0 ) represents the probability of false alarm; D 1 Indicates making a detection decision; H 0 Indicates no signal; V T represents the decision threshold; p(x|H 0 ) means that under the assumption that H 0 The probability of making a detection decision x when is true; x represents the decision that the system can make;

[0063] Among them, V T According to the assumption P(D 1 |H 0 )=P fa Decision; P fa Indicates the false alarm rate; when the judgment result is H 1 When , the probability is expressed as:

[0064]

[0065] The judgment rules are:

[0066]

[0067] Among them, Λ 0 Indicates the set threshold value; H 1 (Λ 0 ∝V T ) indicates the case with a target; ∝ indicates the sign of being proportional to .

[0068] Preferably, the module M4 includes:

[0069] Module M4.1: gridding the three-dimensional imaging area, and treating each grid as a scattering point;

[0070] Module M4.2: Calculate the distance from the scattering point to the transceiver antenna of the ultra-wideband radar; and calculate the time delay between the transceiver signal and the receiver signal according to the distance;

[0071] Module M4.3: corresponding to the pixel data in the one-dimensional range image of the channel echo corresponding to the time delay;

[0072] Module M4.4: traverse all the transceiver channels, accumulate the addressing results to obtain the pixel value of the grid point, and then traverse each of the grid points to obtain the final offset imaging result, that is, the target three-dimensional imaging result;

[0073] In the module M4.4, the three-dimensional imaging result of the target is expressed as follows:

[0074]

[0075] Where I(x, y, z) represents the target three-dimensional imaging result; R jk Indicates the antenna measurement point (x j ,y k ,0) to the imaging point (x,y,z); J represents the maximum value of x movement; K represents the maximum value of y movement; v represents the propagation speed of electromagnetic waves in underground media; Represents the round-trip delay of electromagnetic waves from the imaging point to the corresponding antenna position;

[0076] The R jk The mathematical expression is:

[0077]

[0078] Where x, y, and z represent the three-dimensional coordinates of the target three-dimensional imaging; x j With y k Respectively represent the abscissa and ordinate of the antenna measurement point;

[0079] The calculation amount of the target three-dimensional imaging result is expressed as follows:

[0080] T BP =N g (N a T td +(N a -1)T add )

[0081] Among them, T BP Indicates the amount of calculation; N g Indicates the number of grid points; N a Indicates the number of time delay solutions; T td Indicates the time to solve a delay; T add Indicates the time to perform one addition operation.

[0082] Preferably, in the module M4, artifacts of the target three-dimensional imaging result are reduced by a convolutional neural network, and a target recognition result is output.

[0083] Compared with the prior art, the present invention has the following beneficial effects:

[0084] 1. The three-dimensional fine imaging method proposed in the present invention first detects and locates the two-dimensional imaging results, and then uses the three-dimensional imaging as the back-end network input. Compared with the direct three-dimensional imaging method, it reduces the number of model parameters and improves the hardware processing efficiency, algorithm processing speed and recognition accuracy.

[0085] 2. The present invention adopts a method of coarse imaging first and fine imaging later, so that the system can form a higher resolution three-dimensional image of the target under a limited number of antenna apertures.

[0086] 3. The present invention has strong robustness, which reduces the requirements of the three-dimensional imaging method on radar hardware and the difficulty of radar echo signal processing.

[0087] 4. The present invention uses deep learning technology to intelligently interpret radar images, and converts abstract target radar images into easily understandable information through human posture recognition, thereby realizing intelligent perception of target posture.

[0088] 5. The present invention can improve the resolution of the three-dimensional image of the target, improve the accuracy of neural network recognition, and improve the tolerance of radar hardware through fine imaging under a limited number of antenna apertures. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0090] Figure 1 A flowchart of three-dimensional fine imaging of concealed targets provided by the present invention;

[0091] Figure 2 A schematic diagram of pixel point traversal of the BP imaging algorithm provided by the present invention;

[0092] Figure 3 Schematic diagram of the radar image enhancement convolutional neural network provided by the present invention. DETAILED DESCRIPTION

[0093] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0094] The technical problems to be solved by the present invention are embodied in the following aspects:

[0095] 1. The spatial imaging resolution of concealed targets is limited by the number of antenna apertures, and it is difficult to form a high-resolution three-dimensional image with a limited number of antenna apertures;

[0096] 2. Traditional 3D imaging methods have high requirements on radar RF transceiver hardware, and the quality of radar echo signals cannot meet the requirements of high-resolution 3D imaging;

[0097] 3. Currently, when using deep learning frameworks for radar data imaging and recognition, there are problems such as large number of model parameters, high computational complexity, large data requirements, and low recognition accuracy.

[0098] The present invention adopts a filter to preliminarily filter out stationary clutter, and then adopts a two-dimensional BP algorithm to perform two-dimensional imaging on the azimuth-range plane of the entire detection scene, then adopts a constant false alarm rate detection technology to lock the position of the target in the azimuth-range two-dimensional plane, and then selects a small area where the target is located as a three-dimensional imaging area, and then performs three-dimensional imaging of the three-dimensional imaging area with high grid density, and then inputs the imaging result into a radar fine imaging framework based on a convolutional neural network, so as to realize neural network enhancement of the radar image and suppress false targets in the image.

[0099] The present invention provides a method for three-dimensional fine imaging of a hidden target. The method first uses a two-dimensional BP algorithm for coarse imaging, and then uses a deep neural network to perform fine imaging to achieve target recognition, including:

[0100] Step 1: Use ultra-wideband radar to transmit electromagnetic wave signals, receive hidden target echo signals, and use filtering methods to filter out stationary clutter;

[0101] Step 2: Use the two-dimensional BP imaging algorithm to image the azimuth-distance-height of the entire detection scene to reflect the location, outline and status information of the target;

[0102] Step 3: Perform constant false alarm rate detection and tracking processing on the imaging results to filter out motion clutter, lock the position of the target in the azimuth-range two-dimensional plane, and then select a small area where the target is located as the three-dimensional imaging area;

[0103] Step 4: Perform high-density 3D BP imaging on the target 3D imaging area to obtain the target 3D imaging result as the input of the back-end posture recognition network;

[0104] Step 5: Use convolutional neural network to enhance the radar three-dimensional image, suppress false targets in the radar image, output ideal radar image results and output target recognition results through training.

[0105] Implementation Example 1: Ultra-wideband 3D imaging radar has good spatial information perception capabilities. This method uses radar imaging to obtain the position-distance-height image of the hidden target in 3D space to reflect the position, contour and attitude information of the target. The radar 3D imaging process used in this method is as follows: Figure 1 As shown, including:

[0106] The first step is to use ultra-wideband radar to transmit electromagnetic wave signals, receive hidden target echo signals, and use filtering methods to filter out stationary clutter; wall echo signal elimination mainly includes spatial filtering, principal component analysis, and adaptive estimation and elimination. From the perspective of practical effect, spatial filtering has the strongest robustness. By controlling the transmitting intermediate frequency oscillator and the high-performance spatial frequency filter at the receiving end, the wall echo is effectively suppressed to improve the dynamic range and sensitivity of the system. When the human detection target is close to the wall, a high Q value filter is used to separate the weak target echo signal as much as possible.

[0107] In the second step, a two-dimensional BP imaging algorithm is used to obtain an azimuth-range-altitude image of the entire detection scene to reflect the position, contour and status information of the target; the two-dimensional BP imaging algorithm can be any two-dimensional imaging method, which performs a distance matching rate on the echo signal received by the radar receiving antenna, and then performs an inverse Fourier transform through an inverse Fourier transform to obtain the time delay t of the transmitting and receiving antenna combination, and finally accumulates the coherent addition of the signals to obtain the target function;

[0108] The third step is to perform CFAR detection and tracking processing on the imaging results to filter out motion clutter and lock the position of the target in the azimuth-range two-dimensional plane, and then select a small area where the target is located as the three-dimensional imaging area; in radar signal detection, the probability of the target appearing under certain circumstances is not known in advance, and the loss caused by a missed detection cannot be estimated. Therefore, the Pearson criterion is used in the detection process. Given the false alarm rate detection probability P(D 1 |H 0 )=P fa Under the condition, the detection probability P(D 1 |H 1 ) has the highest probability of occurrence.

[0109] This step is directly related to locking the two-dimensional plane and selecting the three-dimensional imaging area: first, clutter is filtered out according to CFAR detection and tracking processing, which is equivalent to signal preprocessing, and the processed signal is then used for two-dimensional and three-dimensional imaging.

[0110] For a given radar detection system, the input signal and its decision condition are expressed as:

[0111]

[0112] Where x(t) represents the input signal; n(t) represents the system noise; s(t) represents the received signal; t represents time t;

[0113] Because the noise level exceeds the decision threshold V T The probability of false alarm is expressed as:

[0114]

[0115] Among them, P(D 1 |H 0 ) represents the probability of false alarm; D 1 Indicates making a detection decision; H 0 Indicates no signal; V T represents the decision threshold; p(x|H 0 ) means that under the assumption that H 0 The probability of making a detection decision x when is true; x represents the decision that the system can make;

[0116] Among them, V T According to the assumption P(D 1 |H 0 )=R fa Decision; P fa Indicates the false alarm rate; when the judgment result is H 1 When , the mathematical expression of probability is:

[0117]

[0118] Among them, H 1 Indicates that there is a signal;

[0119] The judgment rules are:

[0120]

[0121] Among them, Λ 0 Indicates the set threshold value; H 1 (Λ 0 ∝V T ) means when there is a target; ∝ means the sign of proportional to;

[0122] The fourth step is to perform three-dimensional BP imaging on the target three-dimensional imaging area as the input of the back-end posture recognition network; when applying the BP migration algorithm to process the data, the imaging area is first gridded, and each grid is regarded as a scattering point; then, the distance from the scattering point to the transceiver antenna is calculated, and the time delay between the transceiver signal is calculated based on the distance; the data of the corresponding pixel point in the one-dimensional range image of the corresponding channel echo is addressed based on the time delay; all transceiver channels are traversed, and the addressing results are accumulated to obtain the pixel value of the grid point; each grid point is traversed to obtain the final migration imaging result. Figure 2 Schematic diagram of pixel traversal of BP imaging algorithm in free space.

[0123] Assume that the ground penetrating radar C scan data A(x,y,t) is collected in the xoy plane, A(x j ,y k ,t) is represented by the transmitting and receiving antennas to point (x j ,y k ,0) collected single A scan data, where j = 1, 2, ... J, k = 1, 2, ..., K. Then the imaging result after BP migration processing is

[0124]

[0125] Among them, R jk Indicates the antenna measurement point (x j ,y k ,0) to the imaging point (x,y,z); J represents the maximum value of x movement; K represents the maximum value of y movement; v represents the propagation speed of electromagnetic waves in underground media; Represents the round-trip delay of electromagnetic waves from the imaging point to the corresponding antenna position;

[0126] The R jk The mathematical expression is:

[0127]

[0128] According to the above discussion of the BP migration imaging algorithm, where there are scattering points or diffraction points, the diffraction hyperbolas in each recording track are close to being superimposed in phase, and the pixel value of the corresponding point is very large; on the contrary, where there are no scattering points or diffraction points, the data of each recording track are non-correlatedly superimposed along the diffraction hyperbola, resulting in mutual cancellation, and the total amplitude value after superposition is close to zero, but it is not always zero, which is also the reason for the artificial false image in the imaging result, that is, the artifact. Based on the above principle, the BP migration imaging algorithm can complete the automatic relocation of the reflected wave and the diffraction wave, thereby restoring the true position and geometric characteristics of the underground target.

[0129] Assume that the antenna beam of the shallow ground penetrating radar covers the entire imaging area, and the imaging area is divided into grid points c = X × Y × Z grid points, where X, Y, and Z are the points in elevation, azimuth, and range respectively; the number of radar elevation and azimuth transmitting and receiving antenna sampling points is K, respectively. For each imaging point, its time delay relative to each antenna position should be calculated first, and a total of N a = J×K time delay solutions; then use the time delay to search for J×K echo amplitudes in each A-scan data and accumulate them, performing J×K-1 addition operations. The time for solving a time delay and performing an addition operation is recorded as T td and Tadd , then the amount of computation required to image all points is:

[0130] T BP =N g (N a T td +(N a -1)T add )≈N g N a (T td +T add )

[0131] Among them, N g Indicates the number of grid points; N a Indicates the number of time delay solutions; T td Indicates the time to solve a delay; T add Indicates the time to perform an addition operation;

[0132] According to the above description, the computational complexity of the BP migration imaging algorithm is mainly reflected in the time delay solution of the imaging point relative to each antenna position and the diffraction hyperbola summation operation, and is related to the number of imaging grid points N. g and A-scan data channel number N a The computational cost of three-dimensional imaging is higher due to the larger number of grid divisions and antenna data collection points.

[0133] The fifth step is to enhance the radar three-dimensional image using a convolutional neural network, suppress the false targets in the radar image, output the ideal radar image result, and output the target recognition result through training. The convolutional neural network designed by this method can be divided into seven layers and three parts. The first part is the feature extraction layer, which is composed of the first convolution layer. Its task is to extract features from the complex radar image. The second part is the nonlinear transformation layer, which is composed of the 2nd to 6th convolution layers. It is responsible for processing the features extracted by the feature extraction layer and transmitting them to the output layer. The third part is the output layer, which is composed of the 7th convolution layer. It is responsible for integrating the output data of the nonlinear layer to form an enhanced radar image, that is, the RCS amplitude distribution map in the imaging scene. Compared with the original radar image, the enhanced radar image has the characteristics of low grating side lobes, that is, there are fewer false targets in the image, the image is purer, and the main lobe is sharper. The structure of the convolutional neural network is shown in Table 1.

[0134] Table 1 Convolutional neural network structure

[0135]

[0136]

[0137] In the network training phase, the training sample label pairs are first divided into training set and test set according to a certain ratio, and then the radar image enhancement convolutional network is trained using the training set data, and the performance of the trained network is verified using the test set. The entire training is carried out in a supervised learning manner. The network structure is as follows Figure 3 As shown in Table 1. Using deep neural network output The mean square error, MSE for short, between the label I is used as the loss function. U and V are the number of grids divided in the imaging scene in two directions, and i and j represent the positions of the grids.

[0138]

[0139] The entire deep neural network is trained using the Stochastic Gradient Descent method, or SGD. The initial learning rate is η, and it is gradually decayed during the training process to ensure that there is enough descent rate at the beginning of the training process. In the later stage of training, a lower descent rate is more conducive to network convergence. The above learning parameter settings can be implemented using the PyTorch deep learning framework.

[0140] In the network testing phase, an imaging scene with four point targets is simulated to verify the performance of the trained radar image enhancement convolutional neural network. The four point targets are located at A (3m, 0°, -10°), B (3m, 0°, 10°), C (3m, -10°, 0°), and D (3m, 10°, 0°) in the imaging scene. The RCS amplitude values ​​of targets C and D are twice that of targets A and B. Noise is added to the radar echo according to the signal-to-noise ratio, i.e., signal-to-noise-ratio, referred to as SNR: 0dB, -10dB, and -15dB. The original radar image is obtained, and the original image is enhanced using the constant false alarm rate algorithm, i.e., the Constant False Alarm Rate, referred to as the CFAR algorithm, and the method of this project to obtain the results.

[0141] The present invention also provides a hidden target three-dimensional imaging system, which can be implemented by executing the process steps of the hidden target three-dimensional imaging method, that is, those skilled in the art can understand the hidden target three-dimensional imaging method as a preferred implementation of the hidden target three-dimensional imaging system.

[0142] A concealed target three-dimensional imaging system provided by the present invention comprises:

[0143] Module M1: emits electromagnetic wave signals and receives echo signals from the target;

[0144] Module M2: Based on the echo signal, the azimuth-distance-height of the entire detection scene is imaged to reflect the position, outline and status information of the target;

[0145] Module M3: lock the position of the target in the azimuth-distance two-dimensional plane, and select the range area where the target is located as the three-dimensional imaging area;

[0146] Module M4: obtains a target three-dimensional imaging result according to the three-dimensional imaging area imaging.

[0147] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0148] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for three-dimensional imaging of a concealed target, characterized in that: include: Step S1: transmitting an electromagnetic wave signal and receiving an echo signal of the target; Step S2: Based on the echo signal, the azimuth-distance-height of the entire detection scene is imaged to reflect the position, outline and status information of the target; Step S3: locking the position of the target on the azimuth-distance two-dimensional plane, and selecting the range area where the target is located as the three-dimensional imaging area; Step S4: imaging the three-dimensional imaging area to obtain a target three-dimensional imaging result.

2. The method for three-dimensional imaging of a hidden target according to claim 1, characterized in that: In the step S1: transmitting an electromagnetic wave signal by an ultra-wideband radar; In the step S1, suppressing the wall echo in the echo signal; In the step S2, static clutter of the echo signal is filtered out; In step S2, the azimuth-distance-height of the entire detection scene is imaged by a two-dimensional BP imaging algorithm, namely, a BP algorithm, and the steps include: Step A1: performing distance matching on the echo signal by using the BP algorithm, and then performing inverse Fourier transform on the echo signal to obtain a change result; Step A2: Obtain the time delay t of the transmitting and receiving antenna combination of the ultra-wideband radar, and obtain the target function based on the change result; and form an azimuth-range-altitude image of the detection scene according to the target function.

3. The method for three-dimensional imaging of a hidden target according to claim 2, characterized in that: In step S3, the probability of the target appearing is detected; the judgment condition of the detection result is expressed as follows: Where x(t) represents the input signal; n(t) represents the system noise; s(t) represents the received signal; t represents time t; The probability of false alarm during detection is expressed mathematically as: Where P(D1|H0) represents the probability of false alarm; D1 represents the detection decision; H0 represents no signal; V T represents the decision threshold; p(x|H0) represents the probability of making a detection decision x under the assumption that H0 is true; x represents the decision that the system can make; Among them, V T By assuming that P(D1|H0)=P fa Decision; P fa represents the false alarm rate; when the judgment result is H1, the mathematical expression of the probability is: Among them, H1 means there is a signal; The judgment rules are: Wherein, Λ0 represents the set threshold value; H1(Λ0∝V T ) indicates the case with a target; ∝ indicates the sign of being proportional to .

4. The method for three-dimensional imaging of a concealed target according to claim 3, characterized in that: In the step S4, it includes: Step S4.1: gridding the three-dimensional imaging area, and treating each grid as a scattering point; Step S4.2: Calculate the distance from the scattering point to the transceiver antenna of the ultra-wideband radar; and calculate the time delay between the transceiver signal and the receiver signal according to the distance; Step S4.3: obtaining data of corresponding pixel points in the one-dimensional range image of the channel echo corresponding to the time delay; Step S4.4: traverse all the transceiver channels, accumulate the addressing results to obtain the pixel value of the grid point, and then traverse each of the grid points to obtain the final offset imaging result, that is, the target three-dimensional imaging result; In step S4.4, the target three-dimensional imaging result is expressed as follows: Where I(x, y, z) represents the target three-dimensional imaging result; R jk Indicates the antenna measurement point (x j ,y k ,0) to the imaging point (x,y,z); J represents the maximum value of x movement; K represents the maximum value of y movement; v represents the propagation speed of electromagnetic waves in underground media; Represents the round-trip delay of electromagnetic waves from the imaging point to the corresponding antenna position; The R jk The mathematical expression is: Among them, x, y and z represent the three-dimensional coordinates of the target three-dimensional imaging; j With y k Respectively represent the abscissa and ordinate of the antenna measurement point; The calculation amount of the target three-dimensional imaging result is expressed as follows: T BP =N g (N a T td +(N a -1)T add ) Among them, T BP Indicates the amount of calculation; N g Indicates the number of grid points; N a Indicates the number of time delay solutions; T td Indicates the time to solve a delay; T add Indicates the time to perform one addition operation.

5. The method for three-dimensional imaging of a hidden target according to claim 4, characterized in that: In step S4, artifacts of the target three-dimensional imaging result are reduced by a convolutional neural network, and a target recognition result is output.

6. A three-dimensional imaging system for concealed targets, characterized in that: include: Module M1: emits electromagnetic wave signals and receives echo signals from the target; Module M2: Based on the echo signal, the azimuth-distance-height of the entire detection scene is imaged to reflect the position, outline and status information of the target; Module M3: lock the position of the target in the azimuth-distance two-dimensional plane, and select the range area where the target is located as the three-dimensional imaging area; Module M4: obtains a target three-dimensional imaging result according to the three-dimensional imaging area imaging.

7. The concealed target three-dimensional imaging system according to claim 6, characterized in that: In the module M1: an electromagnetic wave signal is emitted by an ultra-wideband radar; In the module M1, suppressing the wall echo in the echo signal; In the module M2, static clutter of the echo signal is filtered out; In the module M2, the azimuth-distance-height of the entire detection scene is imaged by a two-dimensional BP imaging algorithm, namely, a BP algorithm, and the steps include: Module A1: performing distance matching on the echo signal by using the BP algorithm, and then performing inverse Fourier transform on the echo signal to obtain a change result; Module A2: Obtain the time delay t of the transmitting and receiving antenna combination of the ultra-wideband radar, and obtain the target function based on the change result; and form an azimuth-range-altitude image of the detection scene according to the target function.

8. The concealed target three-dimensional imaging system according to claim 7, characterized in that: In the module M3, the probability of the target appearing is detected; the judgment condition of the detection result is expressed as follows: Where x(t) represents the input signal; n(t) represents the system noise; s(t) represents the received signal; t represents time t; The probability of false alarm during detection is expressed mathematically as: Where P(D1|H0) represents the probability of false alarm; D1 represents the detection decision; H0 represents no signal; V T represents the decision threshold; p(X|H0) represents the probability of making a detection decision x under the assumption that H0 is true; x represents the decision that the system can make; Among them, V T By assuming that P(D1|H0)=P fa Decision; P fa represents the false alarm rate; when the judgment result is H1, the probability is expressed as: The judgment rules are: Wherein, Λ0 represents the set threshold value; H1(Λ0∝V T ) indicates the case with a target; ∝ indicates the sign of being proportional to .

9. The concealed target three-dimensional imaging system according to claim 8, characterized in that: The module M4 includes: Module M4.1: gridding the three-dimensional imaging area, and treating each grid as a scattering point; Module M4.2: Calculate the distance from the scattering point to the transceiver antenna of the ultra-wideband radar; and calculate the time delay between the transceiver signal and the receiver signal according to the distance; Module M4.3: corresponding to the pixel data in the one-dimensional range image of the channel echo according to the time delay; Module M4.4: traverse all the transceiver channels, accumulate the addressing results to obtain the pixel value of the grid point, and then traverse each of the grid points to obtain the final offset imaging result, that is, the target three-dimensional imaging result; In the module M4.4, the three-dimensional imaging result of the target is expressed as follows: Where I(x, y, z) represents the target three-dimensional imaging result; R jk Indicates the antenna measurement point (x j ,y k ,0) to the imaging point (x,y,z); J represents the maximum value of x movement; K represents the maximum value of y movement; v represents the propagation speed of electromagnetic waves in underground media; Represents the round-trip delay of electromagnetic waves from the imaging point to the corresponding antenna position; The R jk The mathematical expression is: Among them, x, y, and z represent the three-dimensional coordinates of the target three-dimensional imaging respectively; x j With y k Respectively represent the abscissa and ordinate of the antenna measurement point; The calculation amount of the target three-dimensional imaging result is expressed as follows: T BP =N g (N a T td +(N a -1)T add ) Among them, T BP Indicates the amount of calculation; N g Indicates the number of grid points; N a Indicates the number of time delay solutions; T td Indicates the time to solve a delay; T add Indicates the time to perform one addition operation.

10. The concealed target three-dimensional imaging system according to claim 9, characterized in that: In the module M4, artifacts of the target three-dimensional imaging result are reduced by a convolutional neural network, and a target recognition result is output.

Images