Electromagnetic inverse scattering imaging method based on planar whirling electromagnetic wave

Abstract

The application discloses an electromagnetic inverse scattering imaging method based on a plane vortex electromagnetic wave, and comprises the following steps: 1, a test platform is built to ensure that the positions of all devices and instruments meet the measurement requirements; 2, an incident wave is generated by a plane vortex electromagnetic wave emission module and is emitted to the inside of an imaging area; 3, evenly distributed transmission antennas around the imaging area sequentially emit electromagnetic waves to the imaging area, after scattering and reflection of the medium, the echo signals are received by an antenna receiving module, and amplitude and phase information are extracted for splicing and combination to obtain a scattering field matrix, a real dielectric constant distribution image is taken as a label, and a data set is constructed; 4, the deep learning imaging module is trained by using the data set to obtain a model having the ability of reconstructing the dielectric constant distribution of an imaging area to be imaged. The application can reconstruct the dielectric constant distribution information in the imaging area to be imaged at a high image resolution, a high anti-environmental interference capability and a fast calculation speed, so that the imaging precision can be improved.

Description

Technical Field

[0001] The present invention belongs to the field of electromagnetic backscattering imaging, and specifically relates to an electromagnetic backscattering imaging method based on plane vortex electromagnetic waves. Background Art

[0002] Plane vortex electromagnetic waves are a special form of vortex electromagnetic waves. Plane vortex electromagnetic waves with different modes can propagate horizontally in the same direction simultaneously, thus solving the problem of different divergence angles of vortex electromagnetic waves of different modes. This lays the foundation for the application of plane vortex electromagnetic waves in electromagnetic inverse scattering imaging.

[0003] Existing electromagnetic inverse scattering imaging methods include Born iteration method, contrast source inversion method, subspace optimization algorithm and deep learning related algorithms.

[0004] The Born iteration method first constructs an initial model by making a preliminary estimate of the dielectric constant distribution within the imaging region. It then uses an iterative method to continuously optimize the initial model to achieve more accurate results. However, the Born iteration method performs poorly under strong scattering conditions and is highly dependent on the initial model. Furthermore, the iterative process is computationally intensive, resulting in long computation times and making it difficult to meet real-time requirements.

[0005] Contrast source inversion is an optimization method based on contrast sources. It gradually optimizes imaging results by minimizing the error between observed and simulated data. This method has a rigorous theoretical foundation and performs well under weak scattering conditions. However, it is sensitive to the choice of initial values ​​and is prone to falling into local optimal solutions under strong scattering conditions. Furthermore, its high computational complexity limits its applicability in real-time imaging scenarios.

[0006] Subspace optimization algorithms reduce computational complexity by transforming the imaging problem into an optimization problem in a low-dimensional subspace. These methods are theoretically mathematically rigorous and can provide superior imaging results under weak scattering conditions. However, subspace optimization algorithms perform less well under strong scattering conditions, and their high requirements for subspace selection and the numerous transfer steps increase algorithm complexity and computational errors.

[0007] Deep learning methods using U-Net offer faster computation and better imaging results than traditional algorithms. However, due to the single dimension of information in traditional plane waves, the dataset constructed from their echoes contains limited information about scatterers within the imaging area, limiting imaging accuracy.

[0008] Another deep learning-based method is to use a generative adversarial network for electromagnetic inverse scattering imaging. The generative adversarial network can generate more realistic permittivity distribution through the adversarial training between the generator and the discriminator, thereby improving the imaging effect. However, the application of the generative adversarial network in the electromagnetic inverse scattering imaging problem has certain limitations. For example, the generative adversarial network model training process is complex, and the quality and quantity of the data set are required to be high. In addition, the stability of the generative adversarial network model is poor, and the mode collapse phenomenon is easy to occur, resulting in lack of diversity and accuracy of the generated results. SUMMARY

[0009] The present application is to solve the above-mentioned deficiencies in the prior art, and proposes a method for electromagnetic inverse scattering imaging based on planar vortex electromagnetic waves, so as to obtain more scattering information inside the object to be imaged by using planar vortex electromagnetic waves with different mode numbers, thereby reconstructing the permittivity distribution in the imaging area with higher imaging accuracy, higher system reliability and anti-environmental interference ability, and faster data processing speed.

[0010] In order to achieve the above-mentioned application purposes, the present application adopts the following technical solutions:

[0011] The method for electromagnetic inverse scattering imaging based on planar vortex electromagnetic waves is characterized in that it is applied to a measurement system composed of a planar vortex electromagnetic wave transmitting module, an antenna receiving module, an electromagnetic inverse scattering imaging network and an object to be imaged; each planar vortex electromagnetic wave transmitting module comprises a plurality of array antenna units, a phase controller, a power divider, a signal generator and a reflector; wherein the reflector is installed on the front surface of the array antenna, and the signal generator is connected to the power divider through the phase controller; each antenna receiving module comprises a radiation unit, a phase and amplitude controller, a low-noise amplifier, a vector signal processor and a signal synchronizer; the electromagnetic inverse scattering imaging method comprises the following steps:

[0012] Step 1, setting the phase of the phase controller in each planar vortex electromagnetic wave transmitting module to , wherein is the orbital angular momentum, n is the number of array antenna units in each planar vortex electromagnetic wave transmitting module, and is initialized to 0.

[0013] Step 2, the signal generator in the jth planar vortex electromagnetic wave transmitting module generates an electromagnetic wave signal with a preset frequency and sends it to the jth phase controller, and the jth phase controller applies The phase shift is obtained and the shifted electromagnetic wave signal is sent to the j-th power divider, so that the j-th power divider distributes the shifted electromagnetic wave signal to each array antenna unit to form The jth single-mode plane vortex electromagnetic wave is emitted toward the center of the object to be imaged, and at the same time, the jth reflector is used to enhance the radiation power of the jth plane vortex electromagnetic wave in the emission direction, thereby forming a uniformly surrounded area with the object to be imaged as the center. Plane vortex electromagnetic wave incident source; j∈[1,M];

[0014] Step 3: The jth incident source emits light toward the center of the object to be imaged, and after being scattered and reflected by the object to be imaged, generates a jth echo signal;

[0015] Step 4: Center the object to be imaged. The radiation units in the antenna receiving modules are evenly arranged in a circle, and receive the jth echo signal at the same time, and send it to each phase and amplitude controller. Each phase and amplitude controller adjusts the phase and amplitude of the received echo signal, and obtains A conditioned echo signal;

[0016] Each low noise amplifier performs noise reduction and amplification on the echo signal after adjustment, and obtains After the amplified echo signal, it is transmitted to the corresponding vector signal processor to extract The amplitude and phase information of the echo signal are extracted After the echo signal emitted by the incident source is obtained, Amplitude and phase information;

[0017] Step 5: The amplitude and phase information are normalized, and the normalized amplitude is assigned to the red channel of the RGB image, the normalized phase information is assigned to the green channel of the RGB image, and the blue channel of the RGB image is padded with 0 to obtain a scattering field image; the real dielectric constant distribution map of the object to be imaged corresponding to the scattering field image is used as the real label to obtain a labeled image sample, and then different objects to be imaged are used to construct a scattering field image. A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ; i∈[1,n], n is the number of images in the single-modal image dataset;

[0018] Step 6: , and adjust the phase of the phase controller in each plane vortex electromagnetic wave transmitting module, and follow the process of steps 2 to 5 to construct A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ;

[0019] Step 7: , and adjust the phase of the phase controller in the plane vortex electromagnetic wave emission module, follow the process of steps 2 to 5 to build A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ;

[0020] Step 8: Construct an electromagnetic inverse scattering imaging network, including three preliminary imaging modules with the same structure but different parameters ( 、 and ), decision layer fusion module, imaging optimization module ;

[0021] Each preliminary imaging module includes: a preliminary encoder and a preliminary decoder; the imaging optimization module Contains an optimized encoder and an optimized decoder;

[0022] Step 8.1, Input the first preliminary imaging module Processed in, get the reconstructed image set , used to construct the loss function , thus the first preliminary imaging module Perform training to obtain the first preliminary imaging module after training ;Will enter After that, the first preliminary reconstructed image set is obtained , and Any i-th image in is denoted as ;

[0023] Step 8.2: Follow the process in step 8.1 and use Training the second preliminary imaging module , get the second preliminary imaging module after training ,Will enter obtain a second preliminary reconstructed image set , and denote any i-th image in as ;

[0024] Step 8.3, train a third preliminary imaging module according to the process of step 8.1, to obtain a trained third preliminary imaging module , input into to obtain a third preliminary reconstructed image set , and denote any i-th image in as ; ;

[0025] Step 8.4, input the images in , and into the decision layer fusion module for processing to obtain a multi-modal fusion image set , and denote any i-th fusion image in as , and the corresponding label as :

[0026] The decision layer fusion module extracts the reconstructed information of , and in the red channel respectively, and correspondingly as the data of the three channels of the i-th multi-modal fusion image , so as to obtain the i-th fusion image , and the size of is ;

[0027] Step 8.5, input into the imaging optimization module for processing to obtain an optimized reconstructed image set , and use it to construct a loss function , so as to train the input imaging optimization module to obtain a trained imaging optimization module ;

[0028] Step 9, the trained electromagnetic inverse scattering imaging model is composed of , , , the decision layer fusion module and , and is used to reconstruct any input to-be-identified scattering field image to obtain a prediction result of the permittivity distribution.

[0029] The electromagnetic backscattering imaging method based on plane vortex electromagnetic waves described in the present invention is also characterized in that step 8.1 includes:

[0030] Step 8.1.1, enter In, first go through After the initial encoder processing, the size is Feature map ,in, for The number of pooling layers in the preliminary encoder, for The number of convolution kernels in the last convolution layer of the preliminary encoder;

[0031] Step 8.1.2, enter It is processed in the preliminary decoder and the output size of the final convolutional layer is The reconstructed image ,in, is the discrete size of the object to be imaged, for The number of upsampling layers in the preliminary decoder of , and ;

[0032] Step 8.1.3: Use formula (1) to construct the first preliminary imaging module The loss function :

[0033] (1)

[0034] Step 8.1.4: First preliminary imaging module Train and minimize using the Adam optimizer , until Until convergence, we get ;

[0035] Furthermore, the step 8.5 includes:

[0036] Step 8.5.1, enter After being processed by the optimized encoder, the size is Feature map ,in, To optimize the number of pooling layers in the encoder, To optimize the number of convolution kernels in the last convolution layer in the encoder;

[0037] Step 8.5.2: Feature Map After being processed by the optimized decoder, the final convolutional layer and a sigmoid activation function output size is The reconstructed image ;

[0038] Step 8.5.3: Use formula (2) to construct the imaging optimization module The loss function :

[0039] (2)

[0040] Step 8.5.4: Optimize the imaging module Train and minimize using the Adam optimizer , until Until convergence, we get ;

[0041] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the electromagnetic inverse scattering imaging method, and the processor is configured to execute the program stored in the memory.

[0042] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program executes the steps of the electromagnetic inverse scattering imaging method when the computer program is executed by a processor.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. This invention introduces plane vortex electromagnetic waves into the electromagnetic inverse scattering imaging problem. Compared with traditional plane waves, plane vortex electromagnetic waves introduce a new degree of freedom, orbital angular momentum. This can capture more diverse information about scatterers and construct a data set containing more scattering information of the objects to be imaged, thereby improving the imaging accuracy of the electromagnetic inverse scattering imaging model.

[0045] 2. This invention utilizes decision-level fusion technology to achieve higher imaging accuracy. By fusing the imaging results of multiple modes of plane vortex electromagnetic waves within the target area, the preliminary imaging results of each single-modal data set can be comprehensively considered, thereby effectively reducing the impact of external factors such as environmental noise and vibration on the imaging accuracy of traditional plane wave electromagnetic inverse scattering problems.

[0046] 3. The preliminary imaging module and imaging optimization module of the present invention are based on an autoencoder network, which uses supervised training to learn the fusion extraction of high-level and low-level spatial features, ensuring the comprehensive capture of spatial information. By combining high-level and low-level spatial features, higher accuracy is achieved in image reconstruction tasks.

[0047] 4. This invention employs a two-stage autoencoder architecture. First, the first stage autoencoder performs preliminary image reconstruction on a single-modality dataset, quickly capturing the image's primary features and structural information for initial reconstruction. A second stage autoencoder then refines the reconstructed image to further extract and optimize detailed features. This two-stage reconstruction approach enables the network to maintain computational efficiency while gradually improving the quality of the reconstructed image, ensuring more accurate and clear static detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the arrangement of the measuring device of the present invention;

[0049] Figure 2a This is a schematic diagram of the network structure of the preliminary imaging module of the present invention;

[0050] Figure 2b Schematic diagram of the network structure of the imaging optimization module of the present invention;

[0051] Figure 2c It is a schematic diagram of the electromagnetic inverse scattering imaging model of the present invention;

[0052] Figure 3a is the real dielectric constant label image of the present invention;

[0053] Figure 3b is the input scattered field image of the present invention;

[0054] Figure 3c is a reconstructed image of each preliminary imaging module and optimized imaging module of the present invention;

[0055] Figure 3d They are frequency histograms of the structural similarity index between the imaging results using the traditional plane wave-based imaging method and the imaging results of the present invention and the true dielectric constant label image. DETAILED DESCRIPTION

[0056] In this embodiment, an electromagnetic inverse scattering imaging method based on planar vortex electromagnetic waves first uses three preliminary imaging modules to generate preliminary reconstruction results for three single-modal data sets. Then, a decision-layer fusion module obtains fused data as input to an imaging optimization module to generate the final detection result. By jointly extracting the scattered field information from the three single-modal data sets and optimizing the reconstruction using the imaging optimization module, the dielectric constant of the object to be imaged is accurately reconstructed. The specific steps are as follows:

[0057] Step 1: Set the phase of the phase controller 1.2 in each plane vortex electromagnetic wave transmitting module 1 to ,in, is the orbital angular momentum, n is the number of array antenna units 1.1 in each plane vortex electromagnetic wave transmitting module 1, initialize ; By making the array antenna unit 1.1 have The equal phase difference of the signal enables the required phase delay to be achieved when the signal is fed to each array antenna element 1.1, thereby generating the required modal number. Plane vortex electromagnetic waves.

[0058] Step 2: The signal generator 1.4 in the j-th plane vortex electromagnetic wave transmitting module 1 generates an electromagnetic wave signal of a preset frequency, which in this embodiment is set to 400 MHz, and sends it to the j-th phase controller 1.2, which applies the electromagnetic wave signal to the j-th phase controller 1.2. The phase shift is obtained by the j-th power divider 1.3, and the j-th power divider 1.3 distributes the shifted electromagnetic wave signal to each array antenna unit 1.1 to form At the same time, the jth reflector 1.5 is used to enhance the radiation power of the jth plane vortex electromagnetic wave in the emission direction, thereby forming a uniformly surrounded area with the object to be imaged as the center. Plane vortex electromagnetic wave incident source; j∈[1, ]; In this embodiment, the test scene is arranged as follows Figure 1 As shown, The distance between the incident source and the center of the object to be imaged is set to 4 meters, with equal angles. The arrangement is carried out in a manner of sequentially emitting plane vortex electromagnetic waves to the object to be imaged, thereby obtaining an echo signal containing internal scattering information of the object to be imaged. In this embodiment, =32.

[0059] Step 3: The jth incident source emits light toward the center of the object to be imaged, and after being scattered and reflected by the object to be imaged, generates the jth echo signal;

[0060] Step 4: Center the object to be imaged. The radiation units 2.1 in the antenna receiving modules 2 are evenly arranged in a circle. The distance between the antenna receiving module 2 and the center of the object to be imaged is set to 4 meters, with equal angles The j-th echo signal is received and sent to each phase and amplitude controller 2.2. Each phase and amplitude controller 2.2 adjusts the phase and amplitude of the received echo signal and obtains The adjusted echo information; this The echo information is generated after a transmitting source emits a plane vortex electromagnetic wave. The antenna receiving modules 2 simultaneously receive the scattered information in all directions of the object to be imaged;

[0061] Each low noise amplifier 2.3 performs noise reduction amplification on the respective adjusted echo signal to obtain After the amplified echo signal, it is transmitted to the corresponding vector signal processor 2.4 to extract The amplitude and phase information of the echo signal are extracted After the echo signal emitted by the incident source is obtained, Amplitude and phase information; after the j-th signal source is transmitted The antenna receiving module 2 obtains The amplitude and phase information is taken as the jth column of the image, thus forming a scattered field matrix.

[0062] Step 5: Add the scattered field matrix The amplitude and phase information are normalized to [0, 255], and the normalized amplitude is assigned to the red channel of the RGB image, the normalized phase information is assigned to the green channel of the RGB image, and the blue channel of the RGB image is padded with 0 to obtain a scattering field image, as shown in Figure 3b As shown in the example; the real dielectric constant distribution map of the object to be imaged corresponding to the scattered field image is used as the real label, such as Figure 3a As shown in the example, a labeled image sample is obtained, and then different objects to be imaged are used to construct A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ; i∈[1,n], n is the number of images in the unimodal image dataset; This example uses the MNIST dataset to train and evaluate the model. The MNIST dataset is a handwritten digit dataset containing 10 categories (digits 0 to 9). 600 images are selected for each digit, totaling 6000 images to form the dataset label. 4000 images (400 per category) are selected for training, 1000 images (100 per category) are used for validation, and another 1000 images (100 per category) are used for testing. The resolution of each image is adjusted to , to correspond to the real and imaginary parts of the dielectric constant distribution after the object to be imaged is discretized. In this embodiment, The image content is a single handwritten number. The digital part corresponds to the relative dielectric constant of the object to be imaged, and the real part is set to 2.0. The background part corresponds to the air, and the real part of the relative dielectric constant is set to 1.0. The imaginary parts of the digital part and the background part are both set to zero. The actual size of the object to be imaged is .

[0063] Step 6: , and adjust the phase of the phase controller 1.2 in each plane vortex electromagnetic wave transmitting module 1, and then follow the process of steps 2 to 5 to construct A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ;

[0064] Step 7: , and adjust the phase of the phase controller 1.2 in the plane vortex electromagnetic wave transmitting module 1, and follow the process of steps 2 to 5 to construct A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ;

[0065] Step 8: Construct an electromagnetic inverse scattering imaging network, including three preliminary imaging modules with the same structure but different parameters ( 、 and ), decision layer fusion module, imaging optimization module ;

[0066] Among them, the network structure of each preliminary imaging module is as follows Figure 2a As shown, they all include: a preliminary encoder and a preliminary decoder; an imaging optimization module It contains an optimized encoder and an optimized decoder. The network structure is as follows Figure 2b As shown;

[0067] Step 8.1, Input the first preliminary imaging module Processed in, get the reconstructed image set , used to construct the loss function , thus the first preliminary imaging module Perform training to obtain the first preliminary imaging module after training ;Will enter After that, the first preliminary reconstructed image set is obtained , and Any i-th image in is denoted as ,like Figure 3c As shown in the example;

[0068] Step 8.1.1, enter In, first go through After two convolutions, batch normalization, and pooling, and then through one convolution layer and one pooling layer (the number of convolution kernels in each convolution layer is 16, 64, and 128 respectively), the size is obtained. Feature map ,in, for The number of pooling layers in the preliminary encoder, in this embodiment, , for The number of convolution kernels in the last convolution layer of the preliminary encoder is, in this embodiment, ; The convolution kernel sizes of all convolution layers used in this embodiment are , the step length is ; Take 3, Take 2.

[0069] Step 8.1.2, enter In the preliminary decoder of , after a combination of 4 convolutions and upsampling (where the number of convolution kernels in each convolution layer is 128, 64, 32, and 16 respectively), the output size of the convolution layer of the last layer is The reconstructed image ,in, is the discrete size of the object to be imaged. In this embodiment, . for The number of upsampling layers in the preliminary decoder of , and .

[0070] Step 8.1.3: Use formula (1) to construct the first preliminary imaging module The loss function :

[0071] (1)

[0072] Step 8.1.4: First preliminary imaging module Training is performed using the Adam optimizer with a learning rate of minimize , until converge, thereby obtaining .

[0073] Step 8.2, according to the process of step 8.1, train the second preliminary imaging module with the network initial parameter settings of , consistent, obtain the trained second preliminary imaging module , input into to obtain a second preliminary reconstructed image set , and record any ith image in as , as shown in the example in ; Figure 3c

[0074] Step 8.3, according to the process of step 8.1, train the third preliminary imaging module with the network initial parameter settings of , consistent, obtain the trained third preliminary imaging module , input into to obtain a third preliminary reconstructed image set , and record any ith image in as , as shown in the example in ; Figure 3c

[0075] Step 8.4, input the images in , and into the decision layer fusion module for processing, to obtain a multi-modal fusion image set , and record any ith fusion image in as , saves the real part information of the preliminary reconstruction results of each preliminary imaging module, and the corresponding label is :

[0076] The decision layer fusion module extracts the reconstruction information of , and in the red channel, respectively, as the data of the three channels of the ith multi-modal fusion image , thereby obtaining the ith fusion image , and the size of is ;

[0077] Step 8.5,​​ Input imaging optimization module Processed in, get the optimized reconstructed image set , and used to construct the loss function , thereby optimizing the input imaging module Perform training to obtain the trained imaging optimization module ;

[0078] Step 8.5.1, enter In the process, it is first processed by the optimized encoder, and then processed by a combination of two convolutions, batch normalization and pooling, and then processed by a convolution layer and a pooling layer to obtain a size of Feature map ,in, To optimize the number of pooling layers in the encoder, in this embodiment, , To optimize the number of convolution kernels in the last convolution layer in the encoder.

[0079] Step 8.5.2, Feature Map After being processed by the optimized decoder, it is first processed by a combination of three convolutions and upsampling, and then the final convolution layer and a sigmoid activation function output size is The reconstructed image .

[0080] Step 8.5.3: Use formula (2) to construct the imaging optimization module The loss function :

[0081] (2)

[0082] Step 8.5.4: Optimize the imaging module Training is performed using the Adam optimizer with a learning rate of minimize , until Until convergence, we get ,Through the decision-making layer fusion of the preliminary imaging results of the preliminary imaging module obtained by training each single-modality dataset, the advantages of the preliminary imaging results of each single-modality dataset are fully utilized to obtain more accurate reconstruction results, such as Figure 3c As shown in the example.

[0083] Step 9: , , , decision layer fusion module and The trained electromagnetic inverse scattering imaging model is composed of Figure 2cThe method is shown and used for reconstructing an arbitrary input to-be-identified scattering field image to obtain a prediction result of the permittivity distribution.

[0084] In the embodiment, an electronic device includes a memory for storing a program supporting a processor to execute the method and the processor configured to execute the program stored in the memory.

[0085] In the embodiment, a computer readable storage medium stores a computer program on the computer readable storage medium, and the computer program is run by a processor to execute the steps of the method.

[0086] In order to verify the important role of the strategy layer fusion of the preliminary reconstruction results of the multi-modal data set in the application, the reconstruction results based on the traditional plane wave and the reconstruction results of the electromagnetic inverse scattering imaging model based on the multi-modal data set are compared qualitatively, and the structure similarity index frequency distribution histogram is used to analyze the pros and cons of the reconstruction results, Figure 3d The structure similarity index frequency distribution histogram of the deep learning algorithm based on the traditional plane wave and the structure similarity index frequency distribution histogram of the application are respectively, and the experimental results show that the imaging results of the application are better than the preliminary imaging results of each single modal data set, and the SSIM frequency distribution histogram of the application is better than the SSIM frequency distribution histogram of the traditional plane wave; the experiment shows that the method of the application can effectively improve the imaging accuracy of the deep learning electromagnetic inverse scattering imaging algorithm based on the traditional plane wave.

Claims

1. An electromagnetic inverse scattering imaging method based on plane vortex electromagnetic waves, characterized in that: is applied by A planar vortex electromagnetic wave transmitting module (1), An antenna receiving module (2), an electromagnetic backscatter imaging network, and a measurement system of an object to be imaged; each planar vortex electromagnetic wave transmitting module (1) comprises: a plurality of array antenna units (1.1), a phase controller (1.2), a power distributor (1.3), a signal generator (1.4), and a reflector (1.5); wherein the reflector (1.5) is installed on the front of the array antenna (1.1), and the signal generator (1.4) is connected to the power distributor (1.3) via a phase controller (1.2); each antenna receiving module (2) comprises: a radiation unit (2.1), a phase and amplitude controller (2.2), a low noise amplifier (2.3), a vector signal processor (2.4), and a signal synchronizer (2.5); and the electromagnetic backscatter imaging method comprises the following steps: Step 1: Set the phase of the phase controller (1.2) in each plane vortex electromagnetic wave transmitting module (1) to ,in, is the orbital angular momentum, n is the number of array antenna units (1.1) in each plane vortex electromagnetic wave transmitting module (1), and the initialization ; Step 2: The signal generator (1.4) in the j-th plane vortex electromagnetic wave transmitting module (1) generates an electromagnetic wave signal of a preset frequency and sends it to the j-th phase controller (1.2), and the j-th phase controller (1.2) applies a The phase shift is obtained by the j-th power divider (1.3), and the shifted electromagnetic wave signal is obtained and sent to the j-th power divider (1.3), so that the j-th power divider (1.3) distributes the shifted electromagnetic wave signal to each array antenna unit (1.1) to form The jth single-mode plane vortex electromagnetic wave is emitted toward the center of the object to be imaged, and at the same time, the radiation power of the jth plane vortex electromagnetic wave in the emission direction is enhanced by the jth reflector (1.5), thereby forming a uniformly surrounded area with the object to be imaged as the center. Plane vortex electromagnetic wave incident source; j∈[1,M]; Step 3: The jth incident source emits light toward the center of the object to be imaged, and after being scattered and reflected by the object to be imaged, generates a jth echo signal; Step 4: Center the object to be imaged. The radiation units (2.1) in the antenna receiving modules (2) are evenly arranged in a circle, and receive the jth echo signal at the same time, and send it to each phase and amplitude controller (2.2). Each phase and amplitude controller (2.2) adjusts the phase and amplitude of the received echo signal, and obtains A conditioned echo signal; Each low noise amplifier (2.3) performs noise reduction amplification on the echo signal after adjustment, and obtains After the amplified echo signal, it is transmitted to the corresponding vector signal processor (2.4) to extract The amplitude and phase information of the echo signal are extracted After the echo signal emitted by the incident source is obtained, Amplitude and phase information; Step 5: The amplitude and phase information are normalized, and the normalized amplitude is assigned to the red channel of the RGB image, the normalized phase information is assigned to the green channel of the RGB image, and the blue channel of the RGB image is padded with 0 to obtain a scattering field image; the real dielectric constant distribution map of the object to be imaged corresponding to the scattering field image is used as the real label to obtain a labeled image sample, and then different objects to be imaged are used to construct a scattering field image. A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ; i∈[1,n], n is the number of images in the single-modal image dataset; Step 6: , and adjust the phase of the phase controller (1.2) in each plane vortex electromagnetic wave transmitting module (1), and then follow the process of steps 2 to 5 to construct A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ; Step 7: , and adjust the phase of the phase controller (1.2) in the plane vortex electromagnetic wave transmitting module (1), and then follow the process of steps 2 to 5 to construct A single-modal image dataset , and Any i-th image in is denoted as , The corresponding true label is recorded as ; Step 8: Construct an electromagnetic inverse scattering imaging network, including three preliminary imaging modules with the same structure but different parameters. 、 and , decision layer fusion module, imaging optimization module ; Each preliminary imaging module includes: a preliminary encoder and a preliminary decoder; the imaging optimization module Contains an optimized encoder and an optimized decoder; Step 8.1, Input the first preliminary imaging module Processed in, get the reconstructed image set , used to construct the loss function , thus the first preliminary imaging module Perform training to obtain the first preliminary imaging module after training ;Will enter After that, the first preliminary reconstructed image set is obtained , and Any i-th image in is denoted as ; Step 8.2: Follow the process in step 8.1 and use Training the second preliminary imaging module , get the second preliminary imaging module after training ,Will enter Get the second preliminary reconstructed image set , and Any i-th image in is denoted as ; Step 8.3: Follow the process in step 8.1 and use Training the third preliminary imaging module , get the third preliminary imaging module after training ,Will enter Get the third preliminary reconstructed image set , and Any i-th image in is denoted as ; Step 8.4, , and The image in the decision layer fusion module is processed to obtain a multimodal fusion image set ,Will Any i-th fused image in is recorded as , and its corresponding label is : The decision layer fusion module extracts 、 and After the reconstruction information of the red channel, the corresponding image is taken as the i-th multimodal fusion image The data of the three channels are obtained to obtain the i-th fused image ,and The size is ; Step 8.5, Input imaging optimization module Processed in, get the optimized reconstructed image set , and used to construct the loss function , thereby optimizing the input imaging module Perform training to obtain the trained imaging optimization module ; Step 9: 、 、 , decision-making layer fusion module and The trained electromagnetic inverse scattering imaging model is constructed and used to reconstruct any input scattering field image to be identified to obtain the prediction result of the dielectric constant distribution.

2. The electromagnetic inverse scattering imaging method based on plane vortex electromagnetic waves according to claim 1, characterized in that: The step 8.1 includes: Step 8.1.1, enter In, first go through After the initial encoder processing, the size is Feature map ,in, for The number of pooling layers in the preliminary encoder, for The number of convolution kernels in the last convolution layer of the preliminary encoder; Step 8.1.2, enter It is processed in the preliminary decoder and the output size of the final convolutional layer is The reconstructed image ,in, is the discrete size of the object to be imaged, for The number of upsampling layers in the preliminary decoder of , and ; Step 8.1.3: Use formula (1) to construct the first preliminary imaging module The loss function : (1) Step 8.1.4: First preliminary imaging module Train and minimize using the Adam optimizer , until Until convergence, we get .

3. The electromagnetic inverse scattering imaging method based on plane vortex electromagnetic waves according to claim 2, characterized in that: The step 8.5 includes: Step 8.5.1, enter After being processed by the optimized encoder, the size is Feature map ,in, To optimize the number of pooling layers in the encoder, To optimize the number of convolution kernels in the last convolution layer in the encoder; Step 8.5.2: The feature map After being processed by the optimized decoder, the final convolutional layer and a sigmoid activation function output size is The reconstructed image ; Step 8.5.3: Use formula (2) to construct the imaging optimization module The loss function : (2) Step 8.5.4: Optimize the imaging module Train and minimize using the Adam optimizer , until Until convergence, we get .

4. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the electromagnetic inverse scattering imaging method according to any one of claims 1 to 3, and the processor is configured to execute the program stored in the memory.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the electromagnetic inverse scattering imaging method according to any one of claims 1 to 3 are executed.

Images