A pure phase hologram generation method and system based on a CCANet network

Abstract

The present application relates to the technical field of computer holography and holographic display, and particularly relates to a pure phase hologram generation method and system based on a CCANet network.The present application comprises constructing a CCANet model, obtaining the input of each branch through a reverse angular spectrum propagation model, the branches including a convolutional neural network branch and a Transformer window attention module branch;obtaining the local feature map of the standardized input image through the convolutional neural network branch;obtaining the global feature map of the standardized input image through the Transformer window attention module branch;performing adaptive weighted residual fusion on the local feature map and the global feature map to generate a predicted pure phase hologram;constructing a loss function using the reconstructed image and the input image to obtain the optimal weight;inputting the target image into the trained CCANet model to obtain the optimal pure phase hologram.The present application reduces the acquisition cost of training data, ensures that the generated hologram strictly conforms to the physical diffraction law, the reconstructed image has no obvious artifacts and noise, and the imaging fidelity is high.

Description

Technical Field

[0001] This invention relates to the field of computational holography and holographic display technology, and in particular to a method and system for generating pure phase holograms based on the CCANet network. Background Technology

[0002] Phase-only holograms (POHs) have become a key technology in fields such as holographic near-eye display, augmented reality (AR), virtual reality (VR), and holographic projection due to their core advantages such as high diffraction efficiency, no conjugate image interference, and high light utilization. They can provide high-fidelity three-dimensional imaging capabilities for immersive optical systems.

[0003] Traditional pure phase hologram generation methods are mainly divided into iterative and non-iterative algorithms. Iterative algorithms, represented by the Gerchberg-Saxton (GS) algorithm, can generate holograms of a certain quality, but require dozens or even hundreds of forward and inverse diffraction iterations, resulting in long computation times and failing to meet the application requirements for real-time generation. Non-iterative algorithms, represented by the random phase method and the quadratic phase method, do not require iteration and have fast computation speed, but the hologram reconstruction quality is poor, with serious loss of detail, artifacts, and noise interference, failing to meet the requirements of high-fidelity imaging. In summary, traditional algorithms have always been unable to overcome the technical bottleneck of the trade-off between computational efficiency and reconstruction quality.

[0004] The development of deep learning technology has provided a new technical path to solve the above problems. Hologram generation methods based on convolutional neural networks (CNNs) can preserve the detailed information of reconstructed images well due to their powerful local feature extraction capabilities. However, they are limited by the receptive field of the convolutional kernel and cannot effectively capture long-distance dependent features of the image, resulting in poor global consistency. Although Transformer-based methods can achieve global feature modeling, they generate extremely high computational costs when directly processing 2K high-resolution images, have stringent hardware requirements, cannot achieve real-time generation, and are difficult to implement in practical application scenarios. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a pure phase hologram generation method and system based on the CCANet network, which solves the problems in the prior art where pure phase hologram generation cannot balance computational efficiency and reconstruction quality, the computational cost of global feature modeling of high-resolution images is large, and the local details and global features cannot be effectively fused.

[0006] To achieve the above and other related objectives, this invention provides a method for generating pure phase holograms based on a CCANet network, comprising:

[0007] S1. Acquire the input image and preprocess the input image to obtain a standardized input image;

[0008] S2. Construct the CCANet model and obtain the input of each branch through the inverse angle spectral propagation model. The branches include the convolutional neural network branch and the Transformer window attention module branch.

[0009] S3. The local feature map of the standardized input image is obtained through the convolutional neural network branch; the global feature map of the standardized input image is obtained through the Transformer window attention module branch; the local feature map and the global feature map are adaptively weighted and fused to generate a predicted pure phase hologram.

[0010] S4. Reconstruct the image from the predicted pure phase hologram using the reverse angular spectral propagation model, and construct a loss function using the reconstructed image and the input image to train the CCANet model parameters and obtain the optimal weights;

[0011] S5. Input the target image into the trained CCANet model to obtain the optimal pure phase hologram;

[0012] S6. Input the optimal pure phase hologram into a phase-type spatial light modulator to obtain a visualized image.

[0013] In one embodiment of the present invention, the preprocessing of the input image in step S1 includes resolution normalization, grayscale conversion, and numerical range standardization of the input image.

[0014] In one embodiment of the present invention, the reverse angular spectrum propagation model in step S2 will input the initial complex amplitude of the holographic plane. Propagated to the image plane, the complex amplitude of the image plane is obtained. The formula is: ,in, Indicates Fourier transform, Indicates the inverse Fourier transform. This represents the initial complex amplitude input to the holographic plane. Represents the complex amplitude of the image plane; Indicates distance from transmission The relevant transfer function, Represented as:

[0015] ,

[0016] in, Indicates the illumination wavelength. and They represent along shaft and Spatial frequency along the axial direction, extracting the complex amplitude of the image plane. Using the phase of the holographic plane and the reverse angular spectrum propagation model, the complex amplitude distribution is obtained:

[0017] The convolutional neural network branch adopts an encoder-decoder structure, consisting of four cascaded downsampling blocks and four cascaded upsampling blocks, with skip connections between the downsampling and upsampling blocks. Each downsampling block is composed of a complex-valued convolutional layer, a batch normalization layer, and a complex-valued ReLU activation function connected sequentially. Similarly, each upsampling block is composed of a complex-valued transposed convolutional layer, a batch normalization layer, and a CReLU activation function connected sequentially. The Transformer window attention module branch is composed of a window attention module, which is built based on the Transformer's non-overlapping window self-attention mechanism.

[0018] In one embodiment of the present invention, the convolutional neural network branching formula in step S3 is:

[0019] ,

[0020] in, This represents the input complex matrix. Denotes the real part of a complex matrix. Represents the imaginary part of a complex matrix. Represents the convolution kernel. Denotes the real part of a complex matrix. Represents the imaginary part of a complex matrix;

[0021] The formula for the window self-attention mechanism of the Transformer window attention module branch is as follows:

[0022] ,

[0023] ,

[0024] ,

[0025] in, , and These represent query, key, and value, respectively. , and This indicates splitting by header and flattening the space; Softmax() represents the transformation of real number weights; and concat() represents merging headers.

[0026] The detailed features obtained from the convolutional neural network branch and the global features obtained from the window attention module branch are fused to obtain features containing both detailed and global information. The formula is as follows:

[0027] ,

[0028] in Indicates the features after fusion. This represents the detailed features obtained from the branches of the convolutional neural network. α represents the global features obtained from the window attention module branch, α represents the weight of the convolutional neural network branch, and β represents the weight of the window attention module branch.

[0029] In one embodiment of the present invention, the formula for the reverse angular spectrum propagation model is:

[0030] ,

[0031] in, Indicates phase extraction, This represents the amplitude constrained on the image plane. Indicates distance from transmission The relevant reverse transfer function, This represents the complex amplitude of the holographic plane obtained through reverse propagation;

[0032] The predicted pure phase hologram is used as the phase distribution of the spatial light modulator (SLM) plane. Combined with unit amplitude, the complex amplitude of the SLM plane is constructed. The complex amplitude distribution of the image plane is obtained through the reverse angular spectral propagation model. The image plane amplitude is extracted to obtain the reconstructed image. A loss function is constructed using the mean square error between the reconstructed image and the input image, with the following formula:

[0033] ,

[0034] in, The total number of pixels in the image. To reconstruct the image, To standardize the input image.

[0035] In one embodiment of the present invention, the phase-type spatial light modulator in step S6 includes: a laser source, an attenuator, a spatial filter, a collimating lens, a polarizer, a beam splitter, and a spatial light modulator arranged in sequence.

[0036] In one embodiment of the present invention, the axes of the laser source, attenuator, spatial filter, collimating lens, polarizer, beam splitter, and spatial light modulator are on a straight line.

[0037] This invention also provides a pure phase hologram generation system based on a CCANet network, comprising:

[0038] An image preprocessing module is used to acquire an input image and preprocess the input image to obtain a standardized input image;

[0039] The model building module is used to build the CCANet model. It obtains the input of each branch through the inverse angle spectral propagation model. The branches include the convolutional neural network branch and the Transformer window attention module branch.

[0040] The feature extraction module is used to obtain local feature maps of the standardized input image through the convolutional neural network branch; obtain global feature maps of the standardized input image through the Transformer window attention module branch; and perform adaptive weighted residual fusion of the local feature maps and global feature maps to generate a predicted pure phase hologram.

[0041] The model training module is used to reconstruct the image from the predicted pure phase hologram using the inverse angular spectral propagation model, and to construct a loss function using the reconstructed image and the input image to train the CCANet model parameters and obtain the optimal weights.

[0042] The hologram generation module is used to input the target image into the trained CCANet model to obtain the optimal pure phase hologram;

[0043] The hologram visualization module is used to input the optimal pure phase hologram into a phase-type spatial light modulator to obtain a visualized image.

[0044] As described above, the pure phase hologram generation method and system based on the CCANet network of the present invention has the following beneficial effects:

[0045] (1) The present invention provides a pure phase hologram generation method based on CCANet network. By constructing a dual-branch CCANet network, it integrates the core advantages of convolutional neural network and Transformer. It accurately extracts local detail features of the image through convolutional branches and effectively models global features and long-distance dependencies through Transformer window attention branches, thus solving the problems of local detail loss and poor global consistency in traditional methods.

[0046] (2) The pure phase hologram generation method based on CCANet network of the present invention embeds the band-limited angle spectrum diffraction physical mode, realizes the self-supervised training of the network, greatly reduces the cost of acquiring training data, and ensures that the generated hologram strictly conforms to the physical diffraction law, the reconstructed image has no obvious artifacts and noise, and the imaging fidelity is high. Attached Figure Description

[0047] Figure 1 This is a flowchart of a pure phase hologram generation method based on the CCANet network according to the present invention.

[0048] Figure 2This is a flowchart of the algorithm for a pure phase hologram generation method based on the CCANet network according to the present invention.

[0049] Figure 3 This is a schematic diagram of a phase-type spatial light modulator structure for a pure phase hologram generation method based on a CCANet network according to the present invention.

[0050] Figure 4 This is a schematic diagram of the structure of a pure phase hologram generation system based on the CCANet network according to the present invention.

[0051] Figure 5 A structural block diagram of a computer-readable storage medium provided in an embodiment of this application. Detailed Implementation

[0052] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0053] This section will describe in detail specific embodiments of the present invention. Preferred embodiments of the present invention are shown in the accompanying drawings. The purpose of the drawings is to supplement the textual description with graphics, so that people can intuitively and vividly understand each technical feature and overall technical solution of the present invention, but they should not be construed as limiting the scope of protection of the present invention.

[0054] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," and "horizontal," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. The terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two elements. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0055] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0056] Please see Figure 1 , Figure 2 , Figure 1 This is a flowchart of a pure phase hologram generation method based on the CCANet network according to the present invention. Figure 2 This is a flowchart illustrating the algorithm for a pure phase hologram generation method based on a CCANet network according to the present invention. The present invention provides a pure phase hologram generation method based on a CCANet network, comprising:

[0057] S1. Acquire the input image and preprocess the input image to obtain a standardized input image;

[0058] S2. Construct the CCANet model and obtain the input of each branch through the inverse angle spectral propagation model. The branches include the convolutional neural network branch and the Transformer window attention module branch.

[0059] S3. The local feature map of the standardized input image is obtained through the convolutional neural network branch; the global feature map of the standardized input image is obtained through the Transformer window attention module branch; the local feature map and the global feature map are adaptively weighted and fused to generate a predicted pure phase hologram.

[0060] S4. Reconstruct the image from the predicted pure phase hologram using the reverse angular spectral propagation model, and construct a loss function using the reconstructed image and the input image to train the CCANet model parameters and obtain the optimal weights;

[0061] S5. Input the target image into the trained CCANet model to obtain the optimal pure phase hologram;

[0062] S6. Input the optimal pure phase hologram into a phase-type spatial light modulator to obtain a visualized image.

[0063] Specifically, the preprocessing described in step S1 includes image resolution normalization, grayscale conversion, and numerical range standardization.

[0064] Specifically, the reverse angular spectrum propagation model described in step S2 will input the initial complex amplitude of the holographic plane. Propagated to the image plane, the complex amplitude of the image plane is obtained. The formula is: ,in, Indicates Fourier transform, Indicates the inverse Fourier transform. This represents the initial complex amplitude input to the holographic plane. Represents the complex amplitude of the image plane; Indicates distance from transmission The relevant transfer function, Represented as:

[0065] ,

[0066] in, Indicates the illumination wavelength. and They represent along shaft and Spatial frequency along the axial direction, extracting the complex amplitude of the image plane. Using the phase of the holographic plane and the reverse angular spectrum propagation model, the complex amplitude distribution is obtained:

[0067] The convolutional neural network branch adopts an encoder-decoder structure, consisting of four cascaded downsampling blocks and four cascaded upsampling blocks, with skip connections between the downsampling and upsampling blocks. Each downsampling block is composed of a complex-valued convolutional layer, a batch normalization layer, and a complex-valued ReLU (CReLU) activation function connected sequentially. Similarly, each upsampling block is composed of a complex-valued transposed convolutional layer, a batch normalization layer, and a CReLU activation function connected sequentially. The Transformer window attention module branch consists of a window attention module, which is built based on the Transformer's non-overlapping window self-attention mechanism.

[0068] Specifically, the formula for the convolutional neural network branch mentioned in step S3 is as follows:

[0069]

[0070] Where W represents the input complex matrix, Wr represents the real part of the complex matrix, Wj represents the imaginary part of the complex matrix, X represents the convolution kernel, Xr represents the real part of the complex matrix, and Xj represents the imaginary part of the complex matrix. The activation function used in the complex-valued neural network proposed in this invention is the complex-valued ReLU (CReLU) function, which can be expressed as:

[0071]

[0072] Relu(·) is the standard activation function. The CRelu activation function activates both the real and imaginary parts. When both the real and imaginary parts are strictly positive or strictly negative, the CRelu activation function satisfies the Cauchy-Riemann equation and can perform complex differential operations. This allows complex-valued neural networks to learn amplitude and phase features simultaneously without the indirect processing required by real-valued neural networks.

[0073] The formula for the window self-attention mechanism of the Transformer window attention module branch is as follows:

[0074]

[0075]

[0076]

[0077] Where Q, K, and V represent query, key, and value, respectively; Qh, Kh, and Vh represent splitting by header and flattening the space; Softmax(·) represents real number weight transformation; and concat(·) represents merging headers.

[0078] The detailed features obtained from the convolutional neural network branch and the global features obtained from the window attention module branch are fused to obtain features that contain both detailed and global information. The formula is as follows:

[0079]

[0080] Where TCCANet represents the fused features, Tccnn represents the detailed features obtained from the convolutional neural network branch, Ttrans represents the global features obtained from the window attention module branch, α represents the weight of the convolutional neural network branch, and β represents the weight of the window attention module branch.

[0081] Specifically, the formula for the reverse angular spectral propagation model described in step S4 is:

[0082] ,

[0083] in, Indicates phase extraction, This represents the amplitude constrained on the image plane. Indicates distance from transmission The relevant reverse transfer function, This represents the complex amplitude of the holographic plane obtained through reverse propagation;

[0084] The predicted pure phase hologram is used as the phase distribution of the spatial light modulator (SLM) plane. Combined with unit amplitude, the complex amplitude of the SLM plane is constructed. The complex amplitude distribution of the image plane is obtained through the reverse angular spectral propagation model. The image plane amplitude is extracted to obtain the reconstructed image. The loss function is constructed using the mean square error between the reconstructed image and the input image, as shown in the formula:

[0085]

[0086] Where N is the total number of pixels in the image, Irec is the reconstructed image, and Iin is the normalized input image.

[0087] Specifically, in step S5, after training is completed, the inverse angular spectral propagation model is used as the input to the branch. The branch includes a convolutional neural network branch and a Transformer window attention module branch, which are used to extract local detail features and global features of the image, respectively. The local feature map and the global feature map are then subjected to adaptive weighted fusion based on the channel dimension, and finally a high-quality pure phase hologram is obtained.

[0088] Specifically, the phase-type spatial light modulator mentioned in step S6 includes:

[0089] Please see Figure 3 , Figure 3 This is a schematic diagram of a phase-type spatial light modulator structure for a pure phase hologram generation method based on a CCANet network according to the present invention. The laser source, attenuator, spatial filter, collimating lens, polarizer, beam splitter, and spatial light modulator are arranged sequentially; the axes of the laser source, attenuator, spatial filter, collimating lens, polarizer, beam splitter, and spatial light modulator are all on a straight line. A visual image can be obtained through a camera.

[0090] Please see Figure 4 , Figure 4 This is a schematic diagram of a pure phase hologram generation system based on a CCANet network according to the present invention. The present invention provides a pure phase hologram generation system based on a CCANet network, comprising:

[0091] An image preprocessing module is used to acquire an input image and preprocess the input image to obtain a standardized input image;

[0092] The model building module is used to build the CCANet model. It obtains the input of each branch through the inverse angle spectral propagation model. The branches include the convolutional neural network branch and the Transformer window attention module branch.

[0093] The feature extraction module is used to obtain local feature maps of the standardized input image through the convolutional neural network branch; obtain global feature maps of the standardized input image through the Transformer window attention module branch; and perform adaptive weighted residual fusion of the local feature maps and global feature maps to generate a predicted pure phase hologram.

[0094] The model training module is used to reconstruct the image from the predicted pure phase hologram using the inverse angular spectral propagation model, and to construct a loss function using the reconstructed image and the input image to train the CCANet model parameters and obtain the optimal weights.

[0095] The hologram generation module is used to input the target image into the trained CCANet model to obtain the optimal pure phase hologram;

[0096] The hologram visualization module is used to input the optimal pure phase hologram into a phase-type spatial light modulator to obtain a visualized image.

[0097] Please see Figure 5 , Figure 5This is a structural block diagram of a computer-readable storage medium provided in an embodiment of this application. The present invention also proposes an electronic device, which includes a processor and a memory. The memory stores program instructions, and the processor executes the program instructions to implement the above-described method for generating a pure phase hologram based on a CCANet network. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The memory may include Random Access Memory (RAM) and may also include Non-Volatile Memory, such as at least one disk storage device. The memory can also be an internal memory of the Random Access Memory (RAM) type. The processor and memory can be integrated into one or more independent circuits or hardware, such as an Application Specific Integrated Circuit (ASIC). It should be noted that when the computer program in the aforementioned memory can be implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, an electronic device, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention.

[0098] On the other hand, embodiments of the present invention also provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the methods provided in the above embodiments, including, for example:

[0099] Acquire the input image and preprocess it to obtain a standardized input image;

[0100] The CCANet model is constructed, and the input of each branch is obtained through the inverse angular spectral propagation model. The branches include a convolutional neural network branch and a Transformer window attention module branch, which are used to extract local detail features and global features of the image, respectively.

[0101] Local feature maps are obtained through a convolutional neural network branch; global feature maps are obtained through a Transformer window attention module branch; the local feature maps and global feature maps are fused to generate a predicted pure phase hologram.

[0102] The image is reconstructed using the reverse angle spectral propagation model, and the CCANet model is trained using a loss function to obtain the optimal weights.

[0103] The target image is input into the trained CCANet model, and a high-quality pure phase hologram is output.

[0104] The generated hologram is input into a phase-type spatial light modulator to obtain a visualized image.

[0105] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. All or some of the modules can be selected according to actual needs to achieve the purpose of this embodiment.

[0106] In summary, the present invention provides a pure phase hologram generation method based on the CCANet network. By applying CCANet to pure phase hologram generation and constructing a dual-branch CCANet hybrid network, it simultaneously achieves accurate extraction of local image details and efficient modeling of global features. Combined with a physical diffraction model, it enables unsupervised training, significantly reducing computational overhead and generating high-quality pure phase holograms.

[0107] The technical principles of the present invention have been described above with reference to specific embodiments. These descriptions are merely for explaining the principles of the invention and should not be construed as limiting the scope of protection of the invention in any way. Based on this explanation, those skilled in the art can readily conceive of other specific embodiments of the invention without inventive effort, and these equivalent variations or substitutions are all included within the scope defined by the claims of this application.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating pure phase holograms based on a CCANet network, characterized in that, include: S1. Acquire the input image and preprocess the input image to obtain a standardized input image; S2. Construct the CCANet model and obtain the input of each branch through the inverse angle spectral propagation model. The branches include the convolutional neural network branch and the Transformer window attention module branch. S3. The local feature map of the standardized input image is obtained through the convolutional neural network branch; the global feature map of the standardized input image is obtained through the Transformer window attention module branch; the local feature map and the global feature map are adaptively weighted and fused to generate a predicted pure phase hologram. S4. Reconstruct the image from the predicted pure phase hologram using the reverse angular spectral propagation model, and construct a loss function using the reconstructed image and the input image to train the CCANet model parameters and obtain the optimal weights; S5. Input the target image into the trained CCANet model to obtain the optimal pure phase hologram; S6. Input the optimal pure phase hologram into a phase-type spatial light modulator to obtain a visualized image.

2. The method for generating a pure phase hologram based on a CCANet network according to claim 1, characterized in that: The preprocessing of the input image in step S1 includes resolution normalization, grayscale conversion, and numerical range standardization of the input image.

3. The method for generating a pure phase hologram based on a CCANet network according to claim 1, characterized in that: The reverse angular spectrum propagation model in step S2 will input the initial complex amplitude of the holographic plane. Propagated to the image plane, the complex amplitude of the image plane is obtained. The formula is: ,in, Indicates Fourier transform, Indicates the inverse Fourier transform. This represents the initial complex amplitude input to the holographic plane. Represents the complex amplitude of the image plane; Indicates distance from transmission The relevant transfer function, Represented as: ， in, Indicates the illumination wavelength. and They represent along shaft and Spatial frequency along the axial direction, extracting the complex amplitude of the image plane. Using the phase of the holographic plane and the reverse angular spectrum propagation model, the complex amplitude distribution is obtained: The convolutional neural network branch adopts an encoder-decoder structure, consisting of four cascaded downsampling blocks and four cascaded upsampling blocks, with skip connections between the downsampling and upsampling blocks. Each downsampling block is composed of a complex-valued convolutional layer, a batch normalization layer, and a complex-valued ReLU activation function connected sequentially. Similarly, each upsampling block is composed of a complex-valued transposed convolutional layer, a batch normalization layer, and a CReLU activation function connected sequentially. The Transformer window attention module branch is composed of a window attention module, which is built based on the Transformer's non-overlapping window self-attention mechanism.

4. The method for generating a pure phase hologram based on a CCANet network according to claim 3, characterized in that: The branching formula of the convolutional neural network in step S3 is as follows: ， in, This represents the input complex matrix. Denotes the real part of a complex matrix. Represents the imaginary part of a complex matrix. Represents the convolution kernel. Denotes the real part of a complex matrix. Represents the imaginary part of a complex matrix; The formula for the window self-attention mechanism of the Transformer window attention module branch is as follows: ， ， ， in, , and These represent query, key, and value, respectively. , and This indicates splitting by header and flattening the space; Softmax() represents the transformation of real number weights; and concat() represents merging headers. The detailed features obtained from the convolutional neural network branch and the global features obtained from the window attention module branch are fused to obtain features containing both detailed and global information. The formula is as follows: ， in Indicates the features after fusion. This represents the detailed features obtained from the branches of the convolutional neural network. α represents the global features obtained from the window attention module branch, α represents the weight of the convolutional neural network branch, and β represents the weight of the window attention module branch.

5. The method for generating a pure phase hologram based on a CCANet network according to claim 4, characterized in that: The formula for the reverse angular spectrum propagation model is: ， in, Indicates phase extraction, This represents the amplitude constrained on the image plane. Indicates distance from transmission The relevant reverse transfer function, This represents the complex amplitude of the holographic plane obtained through reverse propagation; The predicted pure phase hologram is used as the phase distribution of the spatial light modulator (SLM) plane. Combined with unit amplitude, the complex amplitude of the SLM plane is constructed. The complex amplitude distribution of the image plane is obtained through the reverse angular spectral propagation model. The image plane amplitude is extracted to obtain the reconstructed image. A loss function is constructed using the mean square error between the reconstructed image and the input image, with the following formula: ， in, The total number of pixels in the image. To reconstruct the image, To standardize the input image.

6. The method for generating a pure phase hologram based on a CCANet network according to claim 1, characterized in that: The phase-type spatial light modulator in step S6 includes: a laser source, an attenuator, a spatial filter, a collimating lens, a polarizer, a beam splitter, and a spatial light modulator arranged in sequence.

7. The method for generating a pure phase hologram based on a CCANet network according to claim 6, characterized in that: The axes of the laser source, attenuator, spatial filter, collimating lens, polarizer, beam splitter, and spatial light modulator are all on a straight line.

8. A pure phase hologram generation system based on a CCANet network, characterized in that, include: The image preprocessing module (10) is used to acquire the input image and preprocess the input image to obtain a standardized input image; The model building module (20) is used to build the CCANet model. The input of each branch is obtained through the inverse angle spectral propagation model. The branches include the convolutional neural network branch and the Transformer window attention module branch. The feature extraction module (30) is used to obtain the local feature map of the standardized input image through the branch of the convolutional neural network; to obtain the global feature map of the standardized input image through the branch of the Transformer window attention module; and to perform adaptive weighted residual fusion of the local feature map and the global feature map to generate the predicted pure phase hologram. The model training module (40) is used to reconstruct the image from the predicted pure phase hologram through the reverse angular spectral propagation model, and to construct a loss function using the reconstructed image and the input image to train the CCANet model parameters and obtain the optimal weights. The hologram generation module (50) is used to input the target image into the trained CCANet model to obtain the optimal pure phase hologram; The hologram visualization module (60) is used to input the optimal pure phase hologram into the phase-type spatial light modulator to obtain a visualized image.

Images