Orbital angular momentum superposition state identification method based on conditional generative adversarial network

Abstract

The invention discloses an orbital angular momentum superposition state identification method based on a conditional generative adversarial network. Vortex beams which are seriously distorted are effectively recovered and identified through the improved conditional generative adversarial network. A jump connection mechanism is introduced into a network structure of the generator, so that bottom detail features can be reserved, and the generation quality of the image is improved. In addition, a peak signal-to-noise ratio loss function is introduced into the loss function of the generator, and the generated image and the real image are forced to keep higher similarity at the pixel level. On the basis of the improved conditional generative adversarial network, convolution attention modules are introduced into a generator and a discriminator, so that the key information extraction capability of the network is improved. According to the method provided by the invention, the vortex beam which is seriously distorted due to the influence of strong turbulence and long-distance propagation conditions is effectively recovered, and finally relatively high identification accuracy is obtained.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of free space optical communication, and relates to an orbital angular momentum superposition state identification method based on a conditional generative adversarial network. BACKGROUND

[0002] Orbital angular momentum (OAM) is a physical property of a vortex beam with a spiral phase term exp(i l f), where l is the topological charge and f is the azimuthal angle. In theory, a vortex beam has an infinite number of modes, and different OAM modes are orthogonal to each other. Therefore, OAM beams can be used for multiplexing and coding to improve the capacity of a communication system. However, when a vortex beam is transmitted, atmospheric turbulence can cause distortion of the wavefront phase of the vortex beam. Especially in the case of strong turbulence and long-distance transmission, the influence on the vortex beam is more obvious, thereby increasing the difficulty of identifying the OAM mode of the vortex beam.

[0003] Traditional identification methods include interference and diffraction methods. However, these two methods have certain limitations, such as the complexity of experimental setup, strict alignment requirements, and limited flexibility. In recent years, deep learning technology has facilitated its application in the field of optical communication. Since vortex beams with different OAM modes have different physical properties, image classification technology can be used to identify the orbital angular momentum of a vortex beam. This method can automatically extract the features of an OAM beam and allow flexible adjustment of the network structure to adapt to different classification tasks, which has a significant advantage over traditional methods.

[0004] Existing methods for improving the accuracy of OAM mode identification include: at the receiving end, after interference between the distorted OAM beam and the Gaussian beam, the light intensity pattern after interference is identified by a convolutional neural network; at the receiving end, after phase compensation of the distorted OAM beam by GS compensation, the identification is performed by a convolutional neural network; and the distorted OAM beam is restored by a generative adversarial network and then identified. However, the above methods have the problems of high alignment requirements for the Gaussian beam and the received beam, the need for more iteration times for severely distorted OAM beams, resulting in high computational complexity, and low quality of the restored OAM beam light intensity pattern. Therefore, it is of great significance to research a method that has low system complexity, low computational complexity, high recovery quality, and can accurately identify the OAM mode of a vortex beam in the case of strong turbulence and long distance. SUMMARY

[0005] The application provides an orbital angular momentum superposition state recognition method based on a conditional generative adversarial network for recognizing strong turbulence and long-distance propagation OAM superposition states.

[0006] In a first aspect, the embodiments of the application provide an orbital angular momentum superposition state recognition method based on a conditional generative adversarial network, and the specific steps are as follows:

[0007] First, a simulation data set of superposition vortex beams is constructed.

[0008] Then, a conditional generative adversarial network is constructed, including a generator and a discriminator. A skip connection is introduced in the generator, and a convolution attention module is introduced in the generator and the discriminator to improve the network's ability to extract key information.

[0009] The training set in the simulation data set is used to train the conditional generative adversarial network. A peak signal-to-noise ratio loss function is added to the loss function of the generator to force the generated image to have a higher similarity with the real image at the pixel level.

[0010] Finally, a vortex beam test set is obtained and input into the trained conditional generative adversarial network to recognize the orbital angular momentum mode of the superposition vortex beam under different turbulence intensities.

[0011] In one possible implementation, the simulation data set of superposition vortex beams is constructed, and the specific operations are as follows:

[0012] First, Laguerre-Gaussian beams with different orbital angular momentum modes, i.e., superposition OAM beams, are obtained by combining a radial exponential of 0 and a topological charge number of {-4, -1, +2, +5}.

[0013] Random phase screens are constructed based on the power spectrum inversion method and the sub-harmonic method, and a plurality of phase screens are inserted in the propagation path of the superposition OAM beams to simulate an atmospheric turbulence channel, so as to obtain distorted OAM beam intensity graphs.

[0014] By changing the atmospheric turbulence structure constant, the intensity graph data set of the superposition OAM beams transmitted under different turbulence intensities is constructed.

[0015] In a possible implementation, the conditional generative adversarial network is composed of a generator and a discriminator. The generator adopts an encoder-decoder architecture and introduces a skip connection mechanism. The encoder is composed of 4 down-sampling blocks, each of which contains a convolutional layer, a batch normalization layer and a LeakyReLU activation function. The decoder is composed of 4 up-sampling blocks, each of which contains a transposed convolutional layer, a batch normalization layer and a LeakyReLU activation function. The discriminator is composed of 5 convolutional layers and 4 fully connected layers, and a batch normalization layer and a LeakyReLU activation function are introduced after each convolutional layer. In addition, a convolutional attention module is introduced between the 4th down-sampling block and the first up-sampling block of the generator, and the 1st fully connected layer of the discriminator.

[0016] In a possible implementation, the skip connection specifically refers to the fusion of the output feature maps of each down-sampling block of the encoder and the output feature maps of the corresponding level of the decoder by channel dimension splicing. The skip connection adopted is an asymmetric connection, and only part of the sampling blocks are connected. Specifically, the 3rd down-sampling block and the 1st up-sampling block, the 2nd down-sampling block and the 2nd up-sampling block, and the 3rd down-sampling block and the 1st up-sampling block are connected.

[0017] In a possible implementation, step 3 specifically includes:

[0018] The category label and the distorted OAM beam intensity map are taken as the input of the generator, and the OAM beam intensity map generated by the generator, the category label and the real OAM beam intensity map are taken as the input of the discriminator. The loss of the generator includes the adversarial loss, the classification loss and the peak signal-to-noise ratio loss. The loss of the generator is represented as:

[0019]

[0020] wherein, is the adversarial loss of the generator, is the classification loss of the generator, is the loss weight coefficient.

[0021] The peak signal-to-noise ratio loss is:

[0022]

[0023] wherein, is the maximum pixel value of the image, and MSE is the mean square error.

[0024] The loss of the discriminator includes the adversarial loss and the classification loss. The loss of the discriminator is represented as:

[0025]

[0026] wherein, a discriminator adversarial loss, a discriminator classification loss.

[0027] In a possible implementation, after obtaining the recognition result, visualization is performed by using a confusion matrix.

[0028] In a second aspect, an embodiment of the present application provides an orbital angular momentum superposition state recognition device based on a conditional generative adversarial network, and specifically includes:

[0029] A data set construction module is configured to construct a simulation data set of superposition vortex beams.

[0030] The conditional generative adversarial network includes a generator and a discriminator. A skip connection is introduced in the generator, and a convolution attention module is introduced in the generator and the discriminator to improve the extraction capability of the network for key information.

[0031] A training module is configured to train the conditional generative adversarial network based on the simulation data set constructed by the data set construction module. A peak signal-to-noise ratio loss function is added to the loss function of the generator, so as to force the generated image to have a higher similarity with the real image at a pixel level.

[0032] An identification module is configured to use the conditional generative adversarial network trained by the training module to identify the orbital angular momentum mode of the superposition vortex beam under different turbulence intensities.

[0033] In a third aspect, an embodiment of the present application provides an electronic device including a processor and a memory.

[0034] The memory is configured to store a computer program.

[0035] The processor is configured to execute the program stored in the memory, so as to implement the orbital angular momentum superposition state recognition method described in any of the embodiments of the present application.

[0036] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program. When the computer program is executed by a processor, the orbital angular momentum superposition state recognition method described in any of the embodiments of the present application is implemented.

[0037] In a fifth aspect, an embodiment of the present application provides a computer program product including instructions, which, when executed on a computer, cause the computer to perform the orbital angular momentum superposition state recognition method described in any of the embodiments of the present application.

[0038] The present application has the following beneficial effects:

[0039] This invention introduces a skip connection mechanism into the generator's network structure. This skip connection preserves low-level detail features and improves the quality of generated images. Furthermore, a peak signal-to-noise ratio (PSNR) loss function is introduced into the generator's loss function, forcing the generated image to maintain a higher pixel-level similarity to the real image. Based on the improved conditional generative adversarial network, convolutional attention modules are introduced into both the generator and discriminator, which enhances the network's ability to extract key information. This method effectively recovers vortex beams that have undergone severe distortion due to strong turbulence and long-distance propagation conditions, ultimately achieving a high recognition accuracy. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0041] Figure 1 This is a flowchart illustrating the method of an embodiment of the present invention.

[0042] Figure 2 This is a light intensity distribution diagram of the original beam, distorted beam, and restored beam in an embodiment of the present invention.

[0043] Figure 3 The atmospheric turbulence structure constant in this embodiment of the invention is: A schematic diagram of the confusion matrix when the propagation distance is 2km. Detailed Implementation

[0044] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of this application.

[0045] See attached document Figure 1 This invention provides a method for identifying orbital angular momentum superposition states based on conditional generative adversarial networks, comprising the following steps:

[0046] Step 1: Obtain the superimposed OAM beam intensity map after transmission through the atmospheric turbulence channel through numerical simulation as a dataset.

[0047] Step 2: Construct a Conditional Generative Adversarial Network (GAN). Skip connections are introduced into the generator, specifically by concatenating feature maps of the same spatial size from the encoder and decoder along the channel dimension. Convolutional attention modules are introduced into both the generator and discriminator to improve the network's ability to extract key information. A peak signal-to-noise ratio (PSNR) loss function is added to the generator's loss function to force the generated image to maintain a higher pixel-level similarity to the real image.

[0048] Step 3: Input the dataset into the conditional generative adversarial network for training and save the optimal deep learning model.

[0049] Step 4: Obtain the light intensity distribution map again as a test set in the same way as in Step 1, and evaluate the optimal deep learning model obtained in Step 3 using the test set.

[0050] In one possible implementation, step 1 specifically includes:

[0051] The Laguerre-Gaussian (LG) beam is a common type of vortex beam, and its optical field in cylindrical coordinates is expressed as follows:

[0052]

[0053]

[0054] in, The imaginary unit is r, and r is the radial coordinate. For topological load number, Radial index, For transmission distance, It is the azimuth angle. The radius of the waist spot, , Rayleigh length, , The waist radius is For wavelength, For wave number, (.) represents a generalized Laguerre polynomial. This invention selects... =0, ={-4, -1, +2, +5} are used as the basic modes of OAM to form 16 coherent superimposed LG beams.

[0055] This embodiment selects a modified von Karman power spectrum model, and generates random phase screens based on the power spectrum inversion method and the subharmonic method. Several phase screens are inserted into the propagation path of the LG beam to simulate the atmospheric turbulence channel. The phase distribution expression of the subharmonic method is as follows:

[0056]

[0057] in, This is the highest order of the subharmonic network. and They are respectively and Discrete spatial frequency of direction, and For harmonic order index, For pattern index, These are random Fourier coefficients. The beam and turbulence parameters selected in this invention are: =300, =1550nm, =0.06m, =0.0003m, =50m, =200m, =3.

[0058] By changing the simulation conditions, a dataset of superimposed OAM beam intensity maps under different turbulence intensities was obtained multiple times. In this embodiment, the selected turbulence intensity is... = , = and = The selected propagation distance is 2km.

[0059] In one possible implementation, step 2 specifically includes:

[0060] The conditional generative adversarial network described in this invention consists of a generator and a discriminator.

[0061] The generator's network structure mainly consists of an encoder and a decoder. The output feature maps of each layer of the encoder are fused with the corresponding output feature maps of the decoder by concatenating them along the channel dimension. The encoder primarily consists of downsampling blocks, each containing a convolutional layer (4×4 convolution, stride 2, padding 1), a batch normalization layer, and a LeakyReLU activation function. The decoder primarily consists of upsampling blocks, each containing a transposed convolutional layer (4×4 transposed convolution, stride 2, padding 1), a batch normalization layer, and a LeakyReLU activation function. In addition, a peak signal-to-noise ratio (PSNR) loss function is introduced into the generator's loss function.

[0062] The discriminator adopts the network structure of the ACGAN discriminator, which can both determine the authenticity of an image and predict its category. The discriminator consists of 5 convolutional layers (3×3 convolutions, stride 2, padding 1) and 4 fully connected layers, with a batch normalization layer and LeakyReLU activation function introduced after each convolutional layer.

[0063] In one possible implementation, step 3 specifically includes:

[0064] The dataset obtained in step 1 is divided into a training set and a validation set in an 8:2 ratio. The class labels and distorted OAM beam intensity maps are used as input to the generator, while the generated OAM beam intensity map, class labels, and real OAM beam intensity maps are used as input to the discriminator. During training, the generator aims to deceive the discriminator into recognizing the images as real and assigning the generated images to the correct class. The generator's loss includes adversarial loss, classification loss, and peak signal-to-noise ratio (PSNR) loss. The adversarial loss and classification loss can be expressed as:

[0065]

[0066]

[0067] in, Let be the expected value, representing the distribution of the actual data. Medium-sized samples The average value. Let y be the input sample and y be the condition label. This is the output of the generator. represents the adversarial part of the discriminator, and represents the probability that the input sample is real data. For the classification part of the discriminator, it represents the output in a given generated sample. Under the given conditions, the discriminator predicts its label as The probability of peak signal-to-noise ratio (PSNR). The peak signal-to-noise ratio is defined by the mean square error:

[0068]

[0069] in, × Let I be the image size, and K be the intensity maps of the original OAM beam and the generated OAM beam, respectively. The peak signal-to-noise ratio can be defined as:

[0070]

[0071] in, Let be the maximum pixel value of the image. Generally, a higher peak signal-to-noise ratio (PSNR) indicates better image quality. Therefore, the PSNR loss is:

[0072]

[0073] The loss of the generator can be expressed as:

[0074]

[0075] in This is the loss weighting coefficient.

[0076] The goal of the discriminator is to distinguish between real and generated images and correctly classify the image category. The discriminator's loss includes adversarial loss and classification loss, which can be expressed as follows:

[0077]

[0078]

[0079] in, The expected value of sampling from the real data distribution. For real samples, For conditional tags. Given a real sample The discriminator predicts that it belongs to the tag. The conditional probability. Let be the probability that the discriminator classifies a real sample as true. This is the input sample. Samples generated by the generator is the probability that the discriminator classifies a generated sample as true. For a given generated sample The discriminator predicts that it belongs to the tag. The conditional probability. The loss of the discriminator can be expressed as:

[0080]

[0081] In this example, the hyperparameters selected are: batch size of 16, learning rate of 0.0002, and optimizer Adam. After 100 training rounds, the optimal deep learning model is saved.

[0082] In one possible implementation, step 4 specifically includes:

[0083] The test sets are labeled S1, S2, and S3, and then input into the trained optimal model: the distorted OAM beam intensity map and class labels are used as input to the generator, which then recovers the distorted OAM beam. (Appendix) Figure 2 The image displays intensity distribution maps of the original, distorted, and recovered beams. The first column shows the original LG beam, the second the distorted LG beam, and the third the recovered LG beam after processing with a conditional generative adversarial network (GAN). The recovered intensity map, class label, and true intensity map are then input into a discriminator to determine the authenticity of the image and predict the class of the OAM beam. Finally, the test results are visualized using a confusion matrix. Figure 3 for = The diagram illustrates the confusion matrix when z=2km. The vertical axis represents the true mode value, the horizontal axis represents the predicted mode value, the diagonal represents the number of correctly identified images, and the others represent the number of incorrectly identified images.

[0084] From the appendix Figure 2 As can be seen, with the introduction of skip connections and the peak signal-to-noise ratio loss function, the recovered superimposed OAM beam is almost indistinguishable from the original superimposed OAM beam. Table 1 shows the test set recognition accuracy of S1, S2, and S3 with and without the introduction of the convolutional attention module. Under condition S3, the recognition accuracy of the conditional generative adversarial network with the convolutional attention module is improved by 0.375% compared to the condition without the convolutional attention module.

[0085] Table 1

[0086]

[0087] This application also provides a device for identifying orbital angular momentum superposition states based on conditional generative adversarial networks, specifically including:

[0088] Dataset building module: Used to build simulation datasets of superimposed vortex beams.

[0089] Conditional Generative Adversarial Networks (GANs) consist of a generator and a discriminator. Skip connections are introduced into the generator, and convolutional attention modules are incorporated into both the generator and discriminator to enhance the network's ability to extract key information.

[0090] Training Module: The conditional generative adversarial network is trained using a simulated dataset built from the dataset construction module. A peak signal-to-noise ratio (PSNR) loss function is added to the generator's loss function to force generated images to maintain a higher pixel-level similarity to real images.

[0091] Identification Module: The conditional generative adversarial network trained by the training module is used to identify the orbital angular momentum modes of superimposed vortex beams under different turbulence intensities.

[0092] In one possible implementation, the dataset construction module operates as follows:

[0093] First, a Laguerre-Gaussian beam with superimposed orbital angular momentum modes is obtained by combining radial exponent 0 and topological charge numbers {-4,-1,+2,+5}.

[0094] Random phase screens are constructed based on the power spectrum inversion method and the subharmonic method. Several phase screens are inserted into the propagation path of the superimposed OAM beam to simulate the atmospheric turbulence channel and obtain the intensity map of the distorted OAM beam.

[0095] By changing the atmospheric turbulence structure constant, an intensity map dataset of superimposed OAM beams transmitted under different turbulence intensities was constructed.

[0096] In one possible implementation, the conditional generative adversarial network consists of a generator and a discriminator. The generator employs an encoder-decoder architecture and introduces a skip connection mechanism. The encoder consists of four downsampling blocks, each containing a convolutional layer, a batch normalization layer, and a LeakyReLU activation function. The decoder consists of four upsampling blocks, each containing a transposed convolutional layer, a batch normalization layer, and a LeakyReLU activation function. The discriminator consists of five convolutional layers and four fully connected layers, with a batch normalization layer and a LeakyReLU activation function introduced after each convolutional layer. Furthermore, convolutional attention modules are introduced between the fourth downsampling block and the first upsampling block of the generator, and in the first fully connected layer of the discriminator.

[0097] In one possible implementation, the skip connection specifically involves fusing the output feature map of each downsampling block of the encoder with the output feature map of the corresponding layer of the decoder through channel-dimensional concatenation. Unlike the skip connections in the U-Net network, the skip connections in this invention are asymmetric connections, and only a subset of sampling blocks are connected. Specifically, the connections are: the 3rd downsampling block and the 1st upsampling block, the 2nd downsampling block and the 2nd upsampling block, and the 3rd downsampling block and the 1st upsampling block.

[0098] In one possible implementation, the training module is as follows:

[0099] The class labels and distorted OAM beam intensity maps are used as inputs to the generator, while the generated OAM beam intensity map, class labels, and real OAM beam intensity maps are used as inputs to the discriminator. The generator's loss includes adversarial loss, classification loss, and peak signal-to-noise ratio loss. The generator's loss is expressed as:

[0100]

[0101] in, To help the generator combat loss, For the generator classification loss, This is the loss weighting coefficient.

[0102] The peak signal-to-noise ratio loss is:

[0103]

[0104] in, is the maximum pixel value of the image, and MSE is the mean square error.

[0105] The discriminator's loss includes adversarial loss and classification loss. The discriminator's loss is expressed as:

[0106]

[0107] in, To help the discriminator combat loss, The classification loss is for the discriminator.

[0108] In one possible implementation, the device further includes a visualization module for visualizing the recognition results using a confusion matrix.

[0109] This application also provides an electronic device, including a processor and a memory.

[0110] The memory is used to store computer programs.

[0111] When the processor executes a program stored in the memory, it implements any of the methods described in this application.

[0112] In one possible implementation, the electronic device of this application embodiment further includes a communication interface and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus.

[0113] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc.

[0114] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0115] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0116] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0117] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements any of the methods described in this application.

[0118] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the methods described in this application.

[0119] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0120] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0121] The various embodiments in this specification are described in a related manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other.

[0122] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A method for identifying orbital angular momentum superposition states based on conditional generative adversarial networks, characterized in that, The specific steps are as follows: First, a simulation dataset of superimposed vortex beams is constructed; Then, a conditional generative adversarial network is constructed, including a generator and a discriminator; skip connections are introduced in the generator, and convolutional attention modules are introduced in the generator and discriminator to improve the network's ability to extract key information. The training set from the simulation dataset is used to train the conditional generative adversarial network; the peak signal-to-noise ratio loss function is added to the generator's loss function; Finally, the test set of the vortex beam is obtained and input into the trained conditional generative adversarial network to identify the orbital angular momentum modes of the superimposed vortex beam under different turbulence intensities.

2. The method for identifying orbital angular momentum superposition states based on conditional generative adversarial networks according to claim 1, characterized in that, The specific steps for constructing the simulation dataset of superimposed vortex beams are as follows: First, a Laguerre-Gaussian beam with superimposed orbital angular momentum modes is obtained by using a combination of radial exponent 0 and topological charge number {-4,-1,+2,+5} with different orbital angular momentum modes, i.e., superimposed OAM beam. Random phase screens are constructed based on the power spectrum inversion method and the subharmonic method. Several phase screens are inserted into the propagation path of the superimposed OAM beam to simulate the atmospheric turbulence channel and obtain the intensity map of the distorted OAM beam. By changing the atmospheric turbulence structure constant, an intensity map dataset of superimposed OAM beams transmitted under different turbulence intensities was constructed.

3. The method for identifying orbital angular momentum superposition states based on conditional generative adversarial networks according to claim 1, characterized in that, The conditional generative adversarial network consists of a generator and a discriminator; the generator adopts an encoder-decoder architecture and introduces a skip connection mechanism; the encoder consists of 4 downsampling blocks, each of which contains a convolutional layer, a batch normalization layer and a LeakyReLU activation function. The decoder consists of four upsampling blocks, each containing a transposed convolutional layer, a batch normalization layer, and a LeakyReLU activation function; the discriminator consists of five convolutional layers and four fully connected layers, with a batch normalization layer and a LeakyReLU activation function introduced after each convolutional layer; in addition, convolutional attention modules are introduced between the fourth downsampling block and the first upsampling block of the generator, and in the first fully connected layer of the discriminator.

4. The method for identifying orbital angular momentum superposition states based on conditional generative adversarial networks according to claim 3, characterized in that, The skip connection specifically refers to the fusion of the output feature map of each downsampling block of the encoder with the output feature map of the corresponding level of the decoder through channel-dimensional concatenation; the skip connection used is an asymmetric connection, and only some sampling blocks are connected; specifically, the connection is between the 3rd downsampling block and the 1st upsampling block, the 2nd downsampling block and the 2nd upsampling block, and the 3rd downsampling block and the 1st upsampling block.

5. The method for identifying orbital angular momentum superposition states based on conditional generative adversarial networks according to claim 1, characterized in that, Step 3 specifically includes: The class labels and distorted OAM beam intensity maps are used as inputs to the generator, while the generated OAM beam intensity map, class labels, and real OAM beam intensity maps are used as inputs to the discriminator. The generator's loss includes adversarial loss, classification loss, and peak signal-to-noise ratio loss. The generator's loss is expressed as: in, To help the generator combat loss, For the generator classification loss, These are the loss weighting coefficients; The peak signal-to-noise ratio loss is: in, The maximum pixel value of the image is denoted as , and MSE is the mean square error. The discriminant's loss includes adversarial loss and classification loss; the discriminant's loss is expressed as: in, To help the discriminator combat loss, The classification loss is for the discriminator.

6. The method for identifying orbital angular momentum superposition states based on conditional generative adversarial networks according to claim 1, characterized in that, After obtaining the recognition results, they are visualized using a confusion matrix.

7. A device for identifying orbital angular momentum superposition states based on conditional generative adversarial networks, characterized in that, Specifically, it includes: Dataset building module: Used to build simulation datasets of superimposed vortex beams; Conditional Generative Adversarial Networks (GANs) consist of a generator and a discriminator. Introducing skip connections in the generator and convolutional attention modules in both the generator and discriminator improves the network's ability to extract key information. Training module: The conditional generative adversarial network is trained using a simulated dataset built from the dataset construction module; a peak signal-to-noise ratio loss function is added to the generator's loss function; Identification Module: The conditional generative adversarial network trained by the training module is used to identify the orbital angular momentum modes of superimposed vortex beams under different turbulence intensities.

8. An electronic device, characterized in that, Including processor and memory; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the orbital angular momentum superposition state identification method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the orbital angular momentum superposition state identification method according to any one of claims 1-6.

10. A computer program product containing instructions, characterized in that, When it is run on a computer, it causes the computer to execute the orbital angular momentum superposition state identification method according to any one of claims 1-6.

Images