All-optical image generation method based on diffraction neural network and storage medium

Abstract

The invention relates to an all-optical image generation method based on a diffraction neural network and a storage medium. The method comprises the following steps: firstly, performing spatial filtering on an initialized first light field generated by a laser through all-optical sparse coding to form a second light field in a sparse activation mode; and then inputting the second light field into a diffraction neural network generator, and transforming the spatial distribution characteristics of the second light field layer by layer through physical propagation to obtain a target optical representation corresponding to the second light field. And finally, focusing the output target optical representation to an image sensor through a focusing lens device to obtain a target image. According to the invention, an all-optical architecture is adopted for image generation, and an all-optical sparse coding mechanism is introduced in an input stage, so that image generation can be completed in a pure optical path, the problems of energy consumption and delay are reduced, the separability and expression ability of a potential space of a generator are improved, and the speed and efficiency of image generation are improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent optical computing technology, and in particular to an all-optical image generation method and storage medium based on a diffraction neural network. Background Art

[0002] Image generation, a key research area in computer vision, has broad applications in diverse tasks, including artistic creation, autonomous driving, and image restoration. Currently, mainstream image generation methods rely primarily on digital neural networks, with representative examples including generative adversarial networks, variational autoencoders, and generative models based on diffusion processes. While these models have achieved significant results in terms of generation quality and expressiveness, their reliance on electronic computing leads to slow inference speed, high energy consumption, and limited system scalability, making them difficult to meet the practical demands for high speed and low energy consumption.

[0003] In recent years, optical computing, due to its parallelism, passivity, and speed-of-light propagation, has become a key area of ​​focus for building new intelligent computing systems. Traditional optical computing, in order to enhance the generator's ability to express the latent space, introduces digital encoders or latent variable design modules. While this enhances input control, it also compromises the system's all-optical nature, leading to increased energy consumption, increased computational latency, and diminishing the advantages of optical computing in speed and parallelism. Summary of the Invention

[0004] The purpose of this application is to provide an all-optical image generation method and storage medium based on a diffraction neural network, so as to solve the problems of computational delay and high energy consumption when generating images by optical calculation.

[0005] To achieve the above objectives, the present application provides, in a first aspect, an all-optical image generation method based on a diffractive neural network, comprising: Initializing an input first light field, where the first light field is generated by a laser and loaded with a programmable optical device to form a randomly distributed optical representation with a set random distribution of amplitude or phase; spatially filtering the first light field by all-optical sparse coding to form a second light field with a sparse activation pattern; Inputting the second light field into a diffractive neural network generator, transforming the spatial distribution characteristics of the second light field layer by layer through physical propagation, and obtaining a target optical representation corresponding to the second light field; The output target optical representation is focused onto the image sensor through a focusing lens device to obtain a target image.

[0006] A second aspect of the present application provides a computer-readable storage medium, in which a program is stored. The program can be loaded by a processor and execute the above-mentioned all-optical image generation method based on diffraction neural network.

[0007] The beneficial effects of this application are: The present application first performs spatial filtering on the initialized first light field generated by the laser through all-optical sparse coding to form a second light field with a sparse activation pattern. The second light field is then input into the diffraction neural network generator, and the spatial distribution characteristics of the second light field are transformed layer by layer through physical propagation to obtain the target optical representation corresponding to the second light field. Finally, the output target optical representation is focused onto the image sensor through a focusing lens device to obtain the target image. The present application adopts an all-optical architecture, utilizes the diffraction and propagation characteristics of light for image generation, and introduces an all-optical sparse coding mechanism in the input stage, which can enable image generation to be completed under a pure optical path without the intervention of an electronic computing unit, reducing the energy consumption and delay problems in the traditional electronic computing process, and improving the separability and expression ability of the generator's potential space, thereby improving the speed and efficiency of image generation.

[0008] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A schematic flow chart of an all-optical image generation method based on a diffraction neural network provided in an embodiment of the present application; Figure 2 A schematic diagram of the process of all-optical sparse coding provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0011] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly specifying the number of the technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more of the described features. In the description of this application, "plurality" means two or more, unless otherwise specifically qualified. In this application, the word "exemplary" is used to mean "serving as an example, illustration, or illustration." Any embodiment described in this application as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is provided to enable anyone skilled in the art to implement and use the present application. In the following description, details are listed for illustrative purposes. It should be understood that one of ordinary skill in the art will recognize that the present application can be implemented without these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed herein.

[0012] Figure 1 This is a flow chart of a method for generating an all-optical image based on a diffraction neural network provided in an embodiment of the present application. Figure 1 As shown, the method may include steps 101-104, which are described in detail below.

[0013] Step 101: Initialize an input first light field, wherein the first light field is generated by a laser and loaded with a programmable optical device to form a randomly distributed optical representation with a set random distribution of amplitude or phase.

[0014] The first light field is the initial light field of the image generation process, serving as the starting point of the image generation process. Initializing the first light field is the starting step of the all-optical image generation method and provides the foundation for subsequent image generation.

[0015] In one example, a laser can be used to generate a light field. Lasers provide highly coherent and monochromatic light sources. Programmable optical devices, such as spatial light modulators (SLMs), are devices that dynamically control the amplitude and phase of light wavefronts and can load randomly distributed patterns into the light field. By combining the laser-generated light field with the random distribution loaded by the programmable optical device, a randomly distributed optical representation can be constructed, which serves as the starting point for the image generation process.

[0016] Step 102: spatially filter the first light field through all-optical sparse coding to form a second light field with a sparse activation pattern.

[0017] To address the issues of insufficient latent space representation in traditional image generation techniques and their reliance on electronic modules for input encoding, this embodiment introduces an all-optical sparse coding mechanism at the input stage. By optically enhancing the distinguishability and orthogonality between spatial variables in the input random light field, the generator's representational power can be improved without the need for additional telecommunications modulators, thereby maintaining the system's all-optical architecture.

[0018] In the embodiments of the present application, the second light field refers to the light field obtained by subjecting the first light field to all-optical sparse coding. All-optical sparse coding is the process of spatially filtering the first light field. Spatial filtering removes unnecessary information from the first light field, retains key features, and forms a second light field with a sparse activation pattern. The sparse activation pattern helps improve the efficiency and quality of image generation. By subjecting the first light field to all-optical sparse coding to obtain the second light field, the light field can have clearer features and less redundancy, providing higher-quality input for subsequent image generation, making image generation more efficient and accurate.

[0019] Step 103: Input the second light field into the diffractive neural network generator, transform the spatial distribution characteristics of the second light field layer by layer through physical propagation, and obtain the target optical representation corresponding to the second light field.

[0020] In an embodiment of the present application, the diffraction neural network generator is a neural network based on the principle of diffraction, which can simulate the propagation and transformation of light fields in multi-layer media. The second light field can propagate layer by layer in the diffraction neural network generator, and each layer simulates the diffraction and propagation process of the second light field in a specific medium. Through layer-by-layer transformation, the spatial distribution characteristics of the second light field can be gradually adjusted and optimized to generate a target optical representation with specific characteristics. Among them, the target optical representation refers to the light field processed by the diffraction neural network generator, which has the required spatial distribution characteristics and can be used to generate the required target image. The diffraction neural network generator can learn and simulate the complex transformation of the light field, thereby generating an optical representation with specific characteristics, providing high-quality optical signals for the final image generation.

[0021] Step 104: Focus the output target optical representation onto the image sensor through a focusing lens device to obtain a target image.

[0022] In the embodiments of the present application, the focusing lens device is an optical element that can focus a light field onto a specific plane, such as an image sensor. Through the focusing lens device, the target optical representation can be focused onto the image sensor to form a clear image. The image sensor is a device that can convert light signals into electrical signals and is used to capture and record images to generate a target image. The target image refers to the image ultimately formed on the image sensor and is the output result of the all-optical image generation method. It has the required characteristics and quality, and the all-optical image generation process can be completed at this point.

[0023] The all-optical image generation method based on a diffraction neural network in the embodiment of the present application can start from a randomly distributed optical representation and, after processes such as spatial filtering, physical propagation, and focusing, ultimately generate a high-quality target image. This application is a mathematical, physical, and chemical method that adopts an all-optical architecture, utilizing the diffraction and propagation characteristics of light for image generation, and introduces an all-optical sparse coding mechanism at the input stage, which can enable image generation to be completed in a pure optical path without the intervention of an electronic computing unit, reducing the energy consumption and delay problems in the traditional electronic computing process, and improving the separability and expression ability of the generator's potential space, thereby improving the speed and efficiency of image generation.

[0024] Figure 2 This is a flow chart of an all-optical sparse coding process provided in an embodiment of the present application. Figure 2 As shown, in step 102, a magnifying optical device with a set magnification ratio is first used to spatially expand the first light field to obtain multiple magnified regions. The magnifying optical device is used to spatially expand the first light field to facilitate subsequent thinning processing.

[0025] As an example, sampling points in a first light field can be obtained. These sampling points are points that need to be processed and represent key information in the first light field. A magnifying optical device with a set magnification factor is used to magnify the size of the sampling points, resulting in multiple magnified regions, each corresponding to a sampling point. The magnification factor determines the size of the magnified region and can be set as needed.

[0026] Then, a pinhole array is used to retain the target regions of the multiple magnified areas to generate the second light field. The pinhole array is a precisely designed structure that spatially filters the amplified first light field, allowing only the target region of each magnified area to transmit light. For example, the target region can be the center of each magnified area, thus forming a sparse activation pattern.

[0027] As an example, first, the first spatial region coordinates of the multiple magnification areas are calculated according to the magnification of the magnifying optical device. Then, based on the first spatial region coordinates of the multiple magnification areas, the second spatial region coordinates of the target areas of the multiple magnification areas are determined. The first spatial region coordinates refer to the spatial region coordinates of the magnification areas, and the second spatial region coordinates refer to the spatial region coordinates of the target areas. Then, the positions of the pinholes are marked in the mask design drawing of the pinhole array, and the position of each pinhole corresponds to a second spatial region coordinate, and the pinhole array is obtained based on the mask design drawing. The pinhole array is used to selectively transmit a specific area in the light field, namely the target area. Finally, the light of the target areas of the multiple magnification areas is transmitted through the pinhole array to obtain a second light field in a sparse activation mode.

[0028] The all-optical sparse coding process essentially introduces position selectivity and structural sparsity into the two-dimensional light field, making the inputs at different spatial positions more independent of each other and more orthogonal, which helps to improve the subsequent generator's ability to express complex latent spaces and promote diversified image generation. In one example, the sparse coding process can be achieved through an optical combination of a convex lens system and a precision pinhole mask, and is completed entirely in a pure light path without the participation of electronic control or digital computing units. Through sparse coding, the statistical negative correlation between the input sampling points of the diffractive neural network can be enhanced, so that the spatial orthogonality is enhanced, providing a clearer input signal for the diffractive neural network generator.

[0029] In an embodiment of the present application, a diffractive neural network generator is composed of multiple layers of propagation networks, each of which includes a spatial light modulation unit and a free-space propagation unit. Specifically, the diffractive neural network generator is composed of multiple spatial light modulation units alternating with free-space propagation units, forming an optical propagation system similar to a layered neural network.

[0030] In step 103, the second light field is first input into the diffractive neural network generator. Then, for each layer of the propagation network, a two-dimensional phase mask of the second light field is applied to the spatial light modulation unit to modulate the second light field, precisely controlling its amplitude and phase. The modulated second light field is then subjected to interference and diffraction by the free-space propagation unit to obtain a mapping and feature transformation of the second light field corresponding to each layer of the propagation network. This step simulates the natural propagation behavior of light in free space, including interference and diffraction. Finally, based on the mapping and feature transformation of the second light field corresponding to each layer of the propagation network, the spatial distribution characteristics of the second light field are adjusted and optimized layer by layer to obtain a target optical representation of the second light field. The target optical representation is the final output after processing by the diffractive neural network generator and is used to generate the target image. The diffractive neural network generator can adopt various topologies, including linear stacking, parallel connection, and skip connection, and can flexibly adjust the number of modulation units, propagation distance, and mask size according to application requirements.

[0031] In order to significantly reduce the dependence and cost of pixel-by-pixel paired annotation data in the process of training the diffraction neural network generator, the embodiment of the present application adopts an unsupervised learning strategy based on a generative adversarial network. On the one hand, in order to alleviate the gradient response mismatch problem between the derivative neural network generator and the electronic discriminator, the Lipschitz continuity constant (Lipschitz constant) of the electronic discriminator is restricted to a limited range during the training process to stabilize the adversarial training process and improve the convergence efficiency. On the other hand, in order to improve the physical feasibility and energy efficiency of the all-optical generation model, the embodiment of the present application designs a collaborative loss function that integrates the diffraction efficiency constraint based on the energy transmission characteristics of the optical system, aiming to effectively improve the light energy conversion rate and detection sensitivity in the actual system while maintaining the quality of the generated image.

[0032] Based on this, in an embodiment of the present application, the all-optical image generation method further includes the step of training a diffractive neural network generator. First, a training sample set is obtained, which may include multiple first sample light fields. The first sample light fields are the initial sample light fields in the training process and are used to train the diffractive neural network generator, enabling it to learn to generate images similar to real images.

[0033] Next, the first sample light field is initialized and spatially filtered using all-optical sparse coding to form a second sample light field with a sparse activation pattern. This allows the sample light field to have clearer features and less redundancy, providing high-quality input for subsequent image generation. For example, the first sample light field can be spatially expanded using an optical magnification device with a set magnification factor to obtain multiple sample magnification regions. Then, using a set pinhole array, the sample target regions of the multiple sample magnification regions are retained to obtain the second sample light field. The specific process can refer to the method for generating a sparse activation pattern described above.

[0034] Then, the second sample light field is input into the diffraction neural network generator to be trained to obtain the generated image output by the diffraction neural network generator. Through layer-by-layer transformation, the spatial distribution characteristics of the second sample light field can be gradually adjusted and optimized to generate a sample optical representation with specific characteristics, and ultimately generate a high-quality generated image. In one example, the second sample light field can be input into the diffraction neural network generator, and the spatial distribution characteristics of the second sample light field are transformed layer by layer through physical propagation to obtain the sample optical representation corresponding to the second sample light field. Then, the output sample optical representation is focused onto the image sensor through a focusing lens device to obtain the generated image. In this way, the generation process of the full optical sample image can be completed.

[0035] The embodiment of the present application trains the diffraction neural network generator through an unsupervised learning strategy, so the generated image can be sent to the electronic discriminator for evaluation and feedback. The electronic discriminator is composed of an electronic neural network, which is used to judge the distribution consistency between the generated image and the real image, and construct an adversarial loss function based on this. The electronic discriminator obtains the classification result through forward reasoning, and transmits the gradient information of the loss function back to the diffraction neural network generator based on the backpropagation algorithm, which is used to update the phase parameters of the spatial light modulators in each layer of the diffraction neural network to achieve collaborative optimization of the entire system. Since the diffraction neural network generator is an all-optical architecture and the electronic discriminator is digitally implemented, there are natural differences between the two in terms of the scale and continuity of the gradient response. To this end, a continuity constraint strategy is introduced to the electronic discriminator during the training process to limit its Lipschitz continuity constant within a limited range, thereby improving the stability and convergence efficiency of the adversarial training. Specifically, the loss signal of the generated image and the real image can be calculated by adding an electronic discriminator with a continuity preset strategy, and then transmitted back to the diffraction neural network generator to adjust the phase parameters of the diffraction neural network generator.

[0036] Then, the second sample light field is input into the diffraction neural network with adjusted phase parameters, and the loss signal is iterated through the electronic discriminator until the loss signal index meets the set index condition, thereby obtaining a trained diffraction neural network generator. The set index condition can be that the number of iterations reaches a set number, or that the loss signal meets a set loss signal range.

[0037] Specifically, the Wasserstein distance can be used to construct the adversarial loss function between the diffractive neural network generator and the electronic discriminator. The Wasserstein distance is a more stable distance metric suitable for training generative adversarial networks. A gradient penalty term is then added to the adversarial loss function to constrain the Lipschitz constant of the electronic discriminator to within a set range. This range can be customized based on requirements. This ensures that the discriminator output does not fluctuate too drastically, thereby improving training stability.

[0038] During the backward transfer process, the phase parameters of the diffraction neural network generator are first fixed, and the first weight gradient of the electronic discriminator is clipped. The first weight gradient is the weight gradient of the electronic discriminator, so that the Lipschitz constant of the electronic discriminator is within the set range, and the discriminator parameters of the electronic discriminator are updated based on the clipped first weight gradient.

[0039] Then, the discriminator parameters of the electronic discriminator are fixed, and the adversarial-diffractive synergistic loss of the diffractive neural network generator is calculated. The second weight gradient of the diffractive neural network generator is calculated using the backpropagation algorithm. The second weight gradient is the weight gradient of the diffractive neural network generator, and the phase parameters of the diffractive neural network generator are updated based on the second weight gradient. This embodiment proposes a loss function design method for an all-optical image generator. This method combines traditional image quality optimization with diffraction efficiency constraints in physical systems to form an adversarial-diffractive synergistic loss with engineering feasibility.

[0040] Among them, the anti-diffraction collaborative loss function can be obtained by the following formula: ; in, To combat the diffraction synergy loss, is the original diffractive neural network generator adversarial loss, is the energy utilization rate of the diffractive neural network, It is a weighting factor to adjust the balance between image fidelity and light energy utilization.

[0041] Through these steps, the diffraction neural network generator corresponding to the full optical image can be effectively trained, ensuring the stability and convergence of the generator and discriminator during the training process, thereby improving the performance of the diffraction neural network generator, ensuring the high quality of the generated images, and optimizing the efficiency of the diffraction system.

[0042] After training is complete, the electronic discriminator module can be omitted during the inference phase, retaining only the optical input, encoding, generation, and imaging components, enabling image generation based entirely on optical structures. This phase requires no electronic computing unit intervention and can complete image generation in nanoseconds, offering exceptional speed and energy efficiency, making it suitable for high-speed, low-power edge intelligent image synthesis scenarios.

[0043] In step 104, the output target optical representation is first focused onto the image plane of the image sensor, and the intensity distribution of the focused light field is recorded. Focusing the target optical representation using a lens or other optical element ensures that the distribution of the light field on the image sensor is as clear as possible, with the image plane being the target location for light field focus. The light field intensity distribution reflects both the amplitude and phase information of the light field.

[0044] Then, based on the light field intensity distribution, the target optical representation is converted into a target electrical signal. The target electrical signal is an electrical signal generated based on the target optical representation. The target electrical signal represents the intensity information of the light field. In one example, the target electrical signal may also undergo preliminary processing, such as amplification and filtering, to improve signal quality.

[0045] The target electrical signal is then converted to digital form, generating a discretized digital image matrix to produce the initial digital image. Converting continuous electrical signals into discrete digital signals is the foundation of digital image processing. The resulting digital image matrix contains the image's pixel values, with each pixel value corresponding to a photosensitive element on the image sensor. The image is reconstructed using the digital image matrix to produce the initial digital image, where each pixel value corresponds to a point in the image. The resulting initial digital image can be grayscale or color, depending on the type and configuration of the image sensor.

[0046] Finally, the initial digital image undergoes noise reduction and gain preprocessing to produce the target image. For example, Gaussian filtering or median filtering can be applied to reduce noise in the initial digital image. Brightness and contrast can be adjusted to enhance the visual quality of the initial digital image. If the initial digital image is in color, color correction can also be performed to ensure color accuracy and consistency.

[0047] An embodiment of the present application also provides a computer-readable storage medium, which stores a program that can be loaded by a processor and execute any one of the all-optical image generation methods and storage media based on a diffraction neural network in the embodiments of the present application.

[0048] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer program. When all or part of the functions in the above embodiments are implemented by computer program, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above functions can be implemented. In addition, when all or part of the functions in the above embodiments are implemented by computer program, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and saved in the memory of the local device by downloading or copying, or the system of the local device is updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be implemented.

[0049] The above specific examples are used to illustrate the present application, which is only used to help understand the present application and is not intended to limit the present application. For those skilled in the art of the present application, based on the concept of the present application, they can also make some simple deductions, modifications or substitutions.

Claims

1. A method for generating all-optical images based on a diffraction neural network, characterized in that: include: Initializing an input first light field, where the first light field is generated by a laser and loaded with a programmable optical device to form a randomly distributed optical representation with a set random distribution of amplitude or phase; spatially filtering the first light field by all-optical sparse coding to form a second light field with a sparse activation pattern; Inputting the second light field into a diffractive neural network generator, transforming the spatial distribution characteristics of the second light field layer by layer through physical propagation, and obtaining a target optical representation corresponding to the second light field; The output target optical representation is focused onto the image sensor through a focusing lens device to obtain a target image.

2. The all-optical image generation method according to claim 1, characterized in that: The spatial filtering of the first light field by all-optical sparse coding to form a second light field with a sparse activation pattern includes: spatially expanding the first light field using a magnifying optical device having a set magnification to obtain a plurality of magnified regions; The target areas of the plurality of magnified areas are retained by a set pinhole array to obtain the second light field.

3. The all-optical image generation method according to claim 2, wherein: The spatially expanding the first light field by using an amplifying optical device with a set magnification comprises: Acquire sampling points in the first light field; The size of the sampling point is magnified by a magnifying optical device with a set magnification ratio to obtain a plurality of magnified areas, each of which corresponds to one sampling point.

4. The all-optical image generation method according to claim 2, wherein: The method of retaining a plurality of target areas of the magnified areas by setting a pinhole array to obtain the second light field includes: Calculating first spatial region coordinates of the plurality of magnified regions according to the magnification of the magnifying optical device; determining second spatial region coordinates of target regions of the plurality of magnified regions based on the first spatial region coordinates of the plurality of magnified regions; Marking positions of pinholes in a mask design drawing of the pinhole array, and obtaining the pinhole array based on the mask design drawing, wherein the position of each pinhole corresponds to a coordinate in the second spatial region; The second light field is obtained by transmitting light from target areas of the plurality of magnified areas through the pinhole array.

5. The all-optical image generation method according to claim 1, wherein: The diffractive neural network generator is composed of a multi-layer propagation network, each layer of which includes a spatial light modulation unit and a free space propagation unit. The second light field is input into the diffractive neural network generator, and the spatial distribution characteristics of the second light field are transformed layer by layer through physical propagation to obtain a target optical representation corresponding to the second light field, including: inputting the second light field into the diffractive neural network generator; For each layer of the propagation network, a two-dimensional phase mask of the second light field is loaded through the spatial light modulation unit to modulate the second light field; The modulated second light field is subjected to interference and diffraction by the free-space propagation unit to obtain mapping and characteristic transformation of the second light field corresponding to each layer of the propagation network; Based on the mapping and feature transformation of the second light field corresponding to each layer of the propagation network, the target optical representation of the second light field is obtained.

6. The all-optical image generation method according to claim 1, wherein: The method further includes the step of training the diffractive neural network generator, the step comprising: Acquire a training sample set, where the training sample set includes a plurality of first sample light fields; Initializing the first sample light field; spatially filtering the first sample light field by all-optical sparse coding to form a second sample light field with a sparse activation pattern; Inputting the second sample light field into the diffractive neural network generator to be trained to obtain a generated image output by the diffractive neural network generator; Calculating the loss signals of the generated image and the real image by adding an electronic discriminator with a continuity preset strategy, and transmitting the loss signals in reverse to the diffraction neural network generator to adjust the phase parameters of the diffraction neural network generator; The second sample light field is input into the diffraction neural network with adjusted phase parameters, and the loss signal is iterated through the electronic discriminator until the index of the loss signal meets the set index condition, thereby obtaining the trained diffraction neural network generator.

7. The all-optical image generation method according to claim 6, characterized in that: The electronic discriminator adding a continuity preset strategy calculates the loss signals of the generated image and the real image, and transmits the loss signals in reverse to the diffraction neural network generator to adjust the phase parameters of the diffraction neural network generator, including: Using Wasserstein distance to construct an adversarial loss function between the diffraction neural network generator and the electronic discriminator; Adding a gradient penalty term to the adversarial loss function to limit the Lipschitz constant of the electronic discriminator to a set range; During the reverse transfer process, fixing the phase parameter of the diffractive neural network generator, clipping the first weight gradient of the electronic discriminator so that the Lipschitz constant of the electronic discriminator is within a set range, and updating the discriminator parameters of the electronic discriminator based on the clipped first weight gradient; Fixing the discriminator parameters of the electronic discriminator, calculating the adversarial-diffraction cooperative loss of the diffraction neural network generator, calculating the second weight gradient of the diffraction neural network generator through a back propagation algorithm, and updating the phase parameter of the diffraction neural network generator based on the second weight gradient; The anti-diffraction synergy loss function is obtained by the following formula: ； in, is the anti-diffraction synergy loss, is the original diffractive neural network generator adversarial loss, is the energy utilization rate of the diffractive neural network, It is a weighting factor to adjust the balance between image fidelity and light energy utilization.

8. The all-optical image generation method according to claim 6, wherein: Inputting the second sample light field into the diffractive neural network generator to be trained to obtain a generated image output by the diffractive neural network generator includes: Inputting the second sample light field into a diffractive neural network generator, transforming the spatial distribution characteristics of the second sample light field layer by layer through physical propagation, and obtaining a sample optical representation corresponding to the second sample light field; The output optical representation of the sample is focused onto an image sensor through a focusing lens device to obtain the generated image.

9. The all-optical image generation method according to any one of claims 1 to 8, characterized in that: The step of focusing the outputted target optical representation onto an image sensor through a focusing lens device to obtain a target image comprises: Focusing the output target optical representation to the image plane position of the image sensor, and recording the intensity distribution of the light field after focusing; converting the target optical representation into a target electrical signal based on the light field intensity distribution; Performing analog-to-digital conversion on the target electrical signal to generate a discretized digital image matrix to obtain an initial digital image; The initial digital image is subjected to noise reduction and gain preprocessing operations to obtain the target image.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which can be loaded by a processor and execute the all-optical image generation method based on a diffraction neural network according to any one of claims 1 to 9.

Images