Optical imaging system, configuration method and optical imaging method

Abstract

The invention discloses an optical imaging system, a configuration method and an optical imaging method.The optical imaging system comprises an optical module used for modulating incident light from a target object and generating a modulated light signal, a sensing unit used for generating imaging information according to the modulated light signal, and a first model used for conducting frequency domain processing on the imaging information, the first model is used for generating first correction information, and the second model is used for performing spatial domain processing on the first correction information to generate an aberration correction image; the configuration method comprises the steps of obtaining an aberration correction image and a calibration image corresponding to the same target object, determining a loss value based on the aberration correction image and the calibration image, updating the configuration of the first model based on the loss value, and updating the configuration of the second model based on the loss value. Or updating both the configuration of the first model and the configuration of the second model. According to the configuration method provided by the invention, the optical imaging system can adapt to application scenes of multi-modal aberration.

Description

Technical Field

[0001] This application relates to the field of image generation and image enhancement technology, and in particular to an optical imaging system, configuration method and optical imaging method. Background Technology

[0002] Imaging technology serves as a core foundation for fields such as astronomy, microscopic imaging, medical diagnostics, and industrial inspection, and its quality directly impacts the accuracy and reliability of observation results. However, in practical applications, imaging systems often suffer from significant image quality degradation due to various aberrations.

[0003] Existing technologies include schemes that dynamically adjust the wavefront using wavefront sensors and deformable mirrors. While these can correct aberrations to some extent, the systems are complex, costly, and difficult to integrate into portable devices. Furthermore, image post-processing methods such as deconvolution, while simplifying hardware requirements, are sensitive to noise, rely on prior knowledge, and lack generalization ability, making them unsuitable for real-world applications with multimodal aberrations. Especially in compact and lightweight optical imaging systems, the imaging process may involve significant multimodal aberrations such as chromatic aberration, spherical aberration, asymmetric astigmatism, field curvature distortion, and geometric distortion. Existing technologies struggle to adapt to such applications, resulting in poor imaging quality. Summary of the Invention

[0004] One of the purposes of this application is to provide a configuration method for an optical imaging system to solve the technical problem that the optical imaging system configured in the prior art cannot adapt to application scenarios with multimodal aberrations, resulting in poor imaging effect.

[0005] One of the purposes of this application is to provide an optical imaging method.

[0006] One of the purposes of this application is to provide an optical imaging system.

[0007] To achieve one of the above objectives, one embodiment of this application provides a configuration method for an optical imaging system. The optical imaging system includes: an optical module for modulating incident light from a target object to generate a modulated light signal; a sensing unit for generating imaging information based on the modulated light signal; a first model for performing frequency domain processing on the imaging information to generate first correction information; and a second model for performing spatial domain processing on the first correction information to generate an aberration-corrected image. The configuration method includes: obtaining an aberration-corrected image and a calibration image corresponding to the same target object; determining a loss value based on the aberration-corrected image and the calibration image; and updating the configuration of the first model and the configuration of the second model, or updating both the configuration of the first model and the configuration of the second model, based on the loss value.

[0008] Optionally, the optical module includes metasurface optical elements.

[0009] Optionally, the first model is configured to generate a frequency domain representation based on imaging information and perform denoising processing on the frequency domain representation based on a first parameter set to determine the first correction information. The configuration method includes: updating the first parameter set based on the loss value to update the configuration of the first model.

[0010] Optionally, the first model performs linear filtering denoising on the frequency domain representation based on the gain parameter and the bias parameter to determine the first correction information. The configuration method includes: updating the gain parameter, updating the bias parameter, or updating both the gain parameter and the bias parameter based on the loss value to update the configuration of the first model.

[0011] Optionally, the first model performs nonlinear filtering denoising on the frequency domain representation based on weight parameters to determine the first correction information, and the configuration method includes: updating at least one of several weight parameters based on the loss value to update the configuration of the first model.

[0012] Optionally, the first correction information is determined based on the imaging information and the point spread function of the optical module; , It is the frequency domain representation of the first correction information. It is the frequency domain representation of imaging information. It is the frequency domain representation of the point spread function of the optical module. , , , , , These are trainable weight parameters. It is a spatial frequency coordinate.

[0013] Optionally, the second model is constructed based on a convolutional neural network or a generative adversarial neural network, and the configuration method includes: updating the weights or biases of the neural network based on the loss value to update the configuration of the second model.

[0014] Optionally, the configuration method further includes: performing degradation processing on the calibration image based on the point spread function or target aberration of the optical module to generate degradation imaging information; using the degradation imaging information as input to the first model; and determining the aberration correction image by performing frequency domain processing of the first model and spatial domain processing of the second model.

[0015] Optionally, the configuration method includes: obtaining the reference point spread function of the optical module, introducing at least one of process tolerance disturbance or dynamic defocus simulation to the reference point spread function, determining a family of point spread functions, and performing convolution processing on the calibration image according to the family of point spread functions to generate the degradation imaging information.

[0016] Optionally, the configuration method includes one of the following: applying a noise component to the degraded image generated after degradation processing to generate the degraded imaging information; the noise component includes additive noise, multiplicative noise, or both additive and multiplicative noise; performing downsampling processing on the degraded image generated after degradation processing based on a resolution reduction operator to generate the degraded imaging information; the resolution reduction operator is used to perform at least one of pixel region averaging, downsampling after anti-aliasing filtering, and spatial blurring based on convolution kernel; applying a noise component to the degraded image generated after degradation processing, and performing downsampling processing on the degraded image with the applied noise component based on a resolution reduction operator to generate the degraded imaging information.

[0017] To achieve one of the above objectives, one embodiment of this application provides an optical imaging method, comprising: obtaining imaging information, wherein the imaging information is generated based on a modulated light signal, the modulated light signal being generated by modulating incident light from a target object; performing frequency domain processing on the imaging information using a first model to generate first correction information; performing spatial domain processing on the first correction information using a second model to generate an aberration-corrected image; wherein the configuration of the first model and the configuration of the second model are updated and determined based on a loss value, the loss value being determined based on an aberration-corrected image and a calibration image corresponding to the same target object.

[0018] Optionally, the optical imaging method includes: calculating the corresponding frequency domain representation by Fourier transform based on the imaging information; performing linear or nonlinear filtering denoising on the frequency domain representation to obtain a denoised frequency domain representation; and calculating the corresponding spatial domain representation by inverse Fourier transform based on the denoised frequency domain representation to obtain the first correction information.

[0019] To achieve one of the above objectives, one embodiment of this application provides an optical imaging system, comprising: an optical module for modulating incident light from a target object to generate a modulated light signal; a sensing unit for generating imaging information based on the modulated light signal; a first model for performing frequency domain processing on the imaging information to generate first correction information; and a second model for performing spatial domain processing on the first correction information to generate an aberration-corrected image. The configuration of the first model and the configuration of the second model are determined based on a loss value update, wherein the loss value is determined based on an aberration-corrected image and a calibration image corresponding to the same target object.

[0020] Optionally, the optical module includes metasurface optical elements.

[0021] Optionally, the first model is used to denoise the imaging information, and the second model is constructed based on a convolutional neural network or a generative adversarial neural network.

[0022] Compared with existing technologies, the configuration method of the optical imaging system provided in this application corrects the imaging information in the frequency domain and spatial domain by setting a first model and a second model respectively, thereby obtaining an aberration-corrected image at both the global and local levels. This method can cope with the combination of various complex optical variations corresponding to multimodal aberrations. For the configuration of the optical imaging system, by using the aberration-corrected image and calibration image corresponding to the same target object, the configuration of the first model and / or the second model is adjusted in reverse with the calibration image as the optimization target, so that the final generated aberration-corrected image approximates the calibration image, further improving the accuracy of imaging in multimodal aberration scenarios and realizing hybrid optimization in the frequency domain and spatial domain. For different optical modules or different aberration environments, the model configuration most suitable for the current scenario can be determined, which has universal adaptability. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the optical imaging system of this application.

[0024] Figure 2 This is a schematic diagram of the configuration method of the optical imaging system of this application.

[0025] Figure 3 This is a schematic diagram of an imaging information according to this application.

[0026] Figure 4 This is a schematic diagram of a first type of correction information in this application.

[0027] Figure 5 This is a schematic diagram of the configuration method steps in one embodiment of this application.

[0028] Figure 6 This is a schematic diagram of a calibration image according to this application.

[0029] Figure 7 This is a schematic diagram of an aberration-corrected image according to this application.

[0030] Figure 8 This is a schematic diagram of the steps of the optical imaging method of this application. Detailed Implementation

[0031] The present application will now be described in detail with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present application, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of this application.

[0032] It should be noted that the term "comprising" or any other variation thereof is 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 process, method, article, or apparatus.

[0033] Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. There is no necessary correlation between the terms "first," "second," etc.; for example, the inclusion of "second" in one embodiment provided in this application does not necessarily mean that "first" is included in that embodiment, and so on.

[0034] Optical imaging system One embodiment of this application provides an optical imaging system 100, such as... Figure 1 As shown.

[0035] The optical imaging system 100 includes an optical module 11.

[0036] Optical module 11 is used to modulate incident light from target object T0. Optical module 11 is used to generate modulated optical signals.

[0037] The modulation can be at least one of refraction, transmission, diffraction, and reflection. The optical module 11 can be composed of one or more optical elements, which are used to correspondingly implement the functions of refraction, transmission, diffraction, and reflection. In some embodiments, the optical elements can also be used to modulate other properties of the optical signal; for example, the optical elements can be used to modulate the amplitude, intensity, phase, polarization state, frequency, wavelength, or propagation direction of the optical signal.

[0038] The optical imaging system 100 includes a sensing unit 12.

[0039] The sensing unit 12 is used to generate imaging information based on the modulated light signal.

[0040] The sensing unit 12 may include a photodetector array. The sensing unit 12 may be a CMOS (Complementary Metal Oxide Semiconductor) photosensitive element, a CCD (Charge Coupled Device) photosensitive element, or an array of photodetectors.

[0041] The optical module 11 can focus the adjustment light signal onto the side of the sensing unit 12 where the photosensitive device is arranged; the sensing unit 12 then generates imaging information accordingly.

[0042] The imaging information can be in the form of an electrical signal. The sensing unit 12 is used for photoelectric conversion.

[0043] The imaging information can be displayed as an imaging information image, or it can be processed into a frequency domain representation or a spatial domain representation corresponding to the imaging information image.

[0044] The optical imaging system 100 includes a first model 131.

[0045] The first model 131 is used to perform frequency domain processing on the imaging information to generate the first correction information.

[0046] The frequency domain processing can be used to process the frequency characteristics of imaging information; specifically, it can process the frequency information, amplitude information, and phase information of imaging information.

[0047] The first model 131 can be used to correct imaging information in the frequency domain to generate first correction information.

[0048] Before correction, the first model 131 can also be used to generate a frequency domain representation based on the imaging information, and then the frequency domain representation of the imaging information is corrected.

[0049] The first model 131 can be implemented by executing a computer program. The entity corresponding to the first model 131 can be a processor that executes the corresponding computer program, a memory that stores the corresponding computer program, or a combination of the memory and the processor.

[0050] The optical imaging system 100 includes a second model 132.

[0051] The second model 132 is used to perform spatial processing on the first correction information to generate an aberration-corrected image.

[0052] The spatial processing can be used to process the spatial characteristics of imaging information; specifically, it can process the spatial location information, signal intensity information (e.g., grayscale value or RGB component value), gradient information, and edge information of imaging information.

[0053] The second model 132 can be used to correct the first correction information in the spatial domain to generate aberration correction information.

[0054] The aberrations can be chromatic aberration (axial / lateral), spherical aberration, asymmetric astigmatism, field curvature distortion, geometric distortion, or coma. The imaging information processed by the second model 132 can correct at least one of the above aberrations.

[0055] The first correction information can be frequency domain information or spatial domain information. In one embodiment, the first correction information is spatial domain information, and the first model 131 can also be used to transform the obtained frequency domain representation into a spatial domain representation after completing the frequency domain processing of the imaging information to obtain the first correction information.

[0056] The second model 132 can be implemented by executing a computer program. The entity corresponding to the second model 132 can be a processor that executes the corresponding computer program, a memory that stores the corresponding computer program, or a combination of the memory and the processor.

[0057] The configuration of the first model 131 is determined based on the loss value update, which is determined based on the aberration-corrected image and the calibration image corresponding to the same target object.

[0058] The configuration of the first model 131 can also be implemented according to any configuration method of this application.

[0059] The configuration of the second model 132 is determined based on the loss value update, which is determined based on the aberration-corrected image and the calibration image corresponding to the same target object.

[0060] The configuration of the second model 132 can also be implemented according to any configuration method of this application.

[0061] The configuration of the first model 131 and the configuration of the second model 132 are determined by updating the loss value, which is based on the aberration-corrected image and the calibration image corresponding to the same target object.

[0062] The loss value characterizes the difference between the aberration-corrected image and the calibration image.

[0063] The model configuration is updated based on the loss value. Specifically, the calibration image can be used as the optimization target to update the model configuration so that the newly generated aberration-corrected image is similar to the calibration image.

[0064] The aberration-corrected image can be output by the second model 132 after the optical imaging system actually images the target object, or it can be determined by the simulation model corresponding to the optical imaging system based on the target object, or it can be output after the second model 132 corrects the aberration introduced by the calibration image, or it can be output after the second model 132 corrects the calibration image based on the simulation model of the corresponding optical module.

[0065] The aberration-corrected image represents the corrected image obtained after processing by the first model 131 and the second model 132; the aberration-corrected image can be used to reflect the correction capability of the current model.

[0066] A calibration image represents a clear image of the same target object; a calibration image can be used to reflect the true state of the same target object when there are no aberrations.

[0067] During configuration, there can be multiple sets of aberration-corrected images and calibration images, each constituting an aberration-corrected image set and a calibration image set; the images in the aberration-corrected image set and the calibration image set have a corresponding relationship.

[0068] In some embodiments, the configuration of the optical module can also be determined based on the updated loss value.

[0069] In one embodiment, the optical module 11 includes metasurface optical elements 110.

[0070] Metasurface optical elements 110 offer significant advantages in terms of system compactness, lightweight design, and wavefront control freedom, making them suitable for applications in systems or portable devices requiring high precision.

[0071] The metasurface optical element 110 has subwavelength-scale discrete phase modulation characteristics and resonant electromagnetic control mechanism, which leads to significant multimodal aberration problems in the imaging system containing the metasurface optical element 110 during the imaging process.

[0072] In this application, the multimodal aberration problem can be effectively solved by the correction of the first model 131 and the second model 132, thus overcoming the aberration and obtaining clear imaging. In other words, the optical imaging system of this application has a more prominent advantage in scenarios where imaging is performed using metasurface optical elements.

[0073] A metasurface is an artificial layered material with a size smaller than or approximately equal to the wavelength, which can be considered as a two-dimensional counterpart of a metamaterial. The metasurface optical element 110 can achieve the control of the polarization, phase, amplitude, frequency, and propagation mode of electromagnetic waves through surface subwavelength microstructure units (also known as metastructure units), thereby realizing properties such as beam shaping, beam deflection, superlensing, super holography, optical rotation, anti-reflection and anti-reflection.

[0074] The metasurface optical element 110 can be a subwavelength-sized optical element, suitable for current micrometer-scale sensor architectures. At the same time, its fabrication process is compatible with mature semiconductor sensor technology, making it highly practical and economical.

[0075] The metasurface optical element 110 may include a substrate and multiple microstructure units arranged in an array on the substrate, each microstructure unit having a nanostructure at its center and / or vertex. The microstructure units are obtained by dividing the metasurface optical element, with each nanostructure as its center. Each period of nanostructures constitutes a microstructure unit. The microstructure units are close-packed patterns, such as regular squares, regular hexagons, or sectors, with one nanostructure per period, and nanostructures may be located at the vertices and / or center of the microstructure unit. In the case of a regular hexagonal microstructure unit, at least one nanostructure is located at each vertex and center of the hexagon. Similarly, the same applies to sector-shaped and square microstructure units.

[0076] The substrate of the metasurface optical element 110 can be selected from materials with similar refractive indices, such as silicon dioxide, BF33, silicon, and polymethyl methacrylate. The nanostructure can be selected from materials such as single-crystal silicon (c-Si), polycrystalline silicon (p-Si), amorphous silicon (a-Si), compound semiconductors (such as GaN, GaP, GaAs, SiC, etc.), SiO2, TiO2, Si3N4, Nb2O5, Ta2O5, Al, AlSb, AlAs, AlGaAs, AlGaInP, BP, ZnGeP2, and other suitable materials, as well as combinations of the above materials.

[0077] Nanostructures can be configured as polarization-dependent or polarization-independent structures. Depending on the application, the microstructure units can be configured as either polarization-dependent or polarization-independent structures. Examples of polarization-independent structures include cylinders, square prisms, cross-shaped prisms, and square prisms with circular holes. Examples of polarization-dependent structures include elliptical cylinders, rectangular prisms, and hexagonal prisms. Nanostructures can be positive or negative structures. For example, nanostructure shapes include cylinders, hollow cylinders, square prisms, and hollow square prisms.

[0078] The metasurface optical element 110 may also include a protective layer covering the nanostructure. The material of the protective layer may be any material with a low refractive index and absorption coefficient in the visible or near-infrared band, such as: silicon dioxide (SiO2), spin-coated glass (SOG), or polymers such as polymethyl methacrylate (PMMA), polydimethylsiloxane (PDMS), polymethylpentene (PMP), and combinations thereof, or it may be air (i.e., no protective layer).

[0079] The optical module 11 may include one or more metasurface optical elements 110. When the optical module 11 includes multiple metasurface optical elements 110, the multiple metasurface optical elements 110 may have the same or different parameters, and the multiple metasurface optical elements 110 may be spaced apart in the propagation direction of the incident light.

[0080] In one embodiment, the optical imaging system 100 includes at least one lens, at least one metasurface optical element 110, and a sensing unit 12. In a specific embodiment, the optical imaging system 100 includes one lens, one metasurface optical element 110, and one sensing unit 12. In a specific embodiment, the optical imaging system 100 includes multiple lenses, one metasurface optical element 110, and one sensing unit 12. In a specific embodiment, the optical imaging system 100 includes one lens, multiple metasurface optical elements 110, and one sensing unit 12. In a specific embodiment, the optical imaging system 100 includes multiple lenses, multiple metasurface optical elements 110, and one sensing unit 12.

[0081] In a preferred embodiment of this application, starting from the target object, the optical imaging system 100 is sequentially provided with a lens, a metasurface optical element 110, and a sensing unit 12. In this embodiment, the metasurface optical element 110 is disposed on the light-emitting side of the lens, specifically on the rear focal plane of the lens. In other embodiments, the metasurface optical element 110, lens, and sensing unit 12 can be arranged in that order, or multiple lenses and multiple metasurface optical elements 110 can be alternately arranged before the sensing unit 12.

[0082] In some embodiments, the metasurface optical element 110 and the sensing unit 12 can be independently separated or integrally disposed. When integrally disposed, the metasurface optical element 110 and the sensing unit 12 can be combined by means of thermal bonding, plasma bonding, adhesive bonding, etc.; or, the metasurface can be directly fabricated on the sensing unit 12, for example: first, an optical film layer is formed on the light-receiving surface of the semiconductor substrate, and then the optical film layer is etched into a metasurface by photolithography.

[0083] In some embodiments, when the metasurface optical element 110 and the sensing unit 12 are independently and separately configured, the optical imaging system 100 may further include a dielectric layer located between the metasurface optical element 110 and the sensing unit 12. The material of the dielectric layer may be any material with a low refractive index and absorption coefficient in the visible or near-infrared band, such as: silicon dioxide (SiO2), spin-coated glass (SOG), or polymers such as polymethyl methacrylate (PMMA), polydimethylsiloxane (PDMS), polymethylpentene (PMP), and combinations thereof.

[0084] In some embodiments, the optical module 11 or the optical imaging system 100 may further include an illumination element, which may specifically be a light-emitting diode (LED). This allows the optical imaging system to achieve a clearer image of the target object.

[0085] In some embodiments, the optical module 11 or the optical imaging system 100 may further include at least one refractive optical element, at least one diffractive optical element, and / or at least one scattering medium element. The refractive optical element includes, but is not limited to, lenses or prisms made of materials such as optical glass, optical plastics, and optical crystals; the diffractive optical element includes, but is not limited to, two-step or multi-step diffractive optical elements, gratings, Dammann gratings, metasurfaces, holograms, diffusers, phase masks, intensity masks, and spatial light modulators; and the scattering medium element includes, but is not limited to, frosted glass.

[0086] In one embodiment, the first model 131 is used to perform noise reduction processing on the imaging information.

[0087] The object of denoising can be imaging information or a frequency domain representation generated from the imaging information. The first correction information generated by denoising can be an image, a frequency domain representation, or a spatial domain representation corresponding to the imaging information.

[0088] In one specific embodiment, the first model 131 is used to generate a frequency domain representation based on imaging information, and to perform noise reduction processing on the frequency domain representation based on a first parameter set to determine first correction information.

[0089] The denoising process can be either linear filtering or nonlinear filtering. Specifically, the nonlinear filtering can be Wiener filtering, or an adaptive Wiener filtering that is an improvement on Wiener filtering.

[0090] In one specific embodiment, the first model 131 performs linear filtering denoising on the frequency domain representation based on gain parameters and bias parameters to determine first correction information.

[0091] In one specific embodiment, the first model 131 performs nonlinear filtering denoising on the frequency domain representation based on weight parameters to determine first correction information. The weight parameters may be model parameters of the first model 131.

[0092] The first model 131 can generate a frequency domain representation based on the imaging information through Fourier transform.

[0093] The first model 131 can obtain the first correction information by generating a spatial domain representation based on the denoised frequency domain representation through inverse Fourier transform.

[0094] In one embodiment, the second model can be built based on a convolutional neural network, a generative adversarial neural network, or any other effective neural network architecture, without limitation on the specific network structure.

[0095] In one specific embodiment, the second model 132 is constructed based on a convolutional neural network.

[0096] A convolutional neural network may include an encoder and a decoder. The encoder may include several convolutional layers and downsampling layers; the decoder may include several upsampling layers and convolutional layers. There may be skip connections between the encoder and the decoder.

[0097] In convolutional neural networks (CNNs), convolutional layers can be followed by normalization layers and activation function layers; ReLU can be used as the activation function. CNNs can also include residual learning blocks to learn the residuals between the inputs of the CNN.

[0098] The second model 132 can be a typical convolutional neural network or a convolutional neural network that is modified as needed.

[0099] In one specific embodiment, the second model 132 is constructed based on a generative adversarial neural network.

[0100] A generative adversarial neural network (GAN) may include a generator and a discriminator. The generator may include an encoding part and a decoding part. There may be skip connections between the encoding part and the decoding part. The discriminator is used to determine the authenticity of the generator's output information. The discriminator may include convolutional layers, normalization layers, and activation functions.

[0101] Configuration method of optical imaging system One embodiment of this application provides a configuration method for an optical imaging system, such as... Figure 2 As shown.

[0102] Combination Figure 1 The optical imaging system 100 includes an optical module 11. The optical module 11 is used to modulate incident light from the target object T0 to generate a modulated optical signal.

[0103] The optical module 11 can be configured based on any of the technical solutions in this application.

[0104] The optical imaging system 100 includes a sensing unit 12. The sensing unit 12 is used to generate imaging information based on a modulated light signal.

[0105] The sensing unit 12 can be configured based on any of the technical solutions in this application.

[0106] The optical imaging system 100 includes a first model 131. The first model 131 is used to perform frequency domain processing on the imaging information to generate first correction information.

[0107] The first model 131 can be configured based on any technical solution of this application.

[0108] The optical imaging system 100 includes a second model 132. The second model 132 is used to perform spatial processing on the first correction information to generate an aberration-corrected image.

[0109] The second model 132 can be configured based on any technical solution of this application.

[0110] Combination Figure 2 As shown, the configuration method of the optical imaging system includes at least one of the following steps.

[0111] Step S1: Obtain aberration-corrected image and calibration image corresponding to the same target object.

[0112] Step S2: Determine the loss value based on the aberration-corrected image and the calibration image.

[0113] Step S3: Based on the loss value, update the configuration of the first model, update the configuration of the second model, or update both the configuration of the first model and the configuration of the second model.

[0114] The calibration image can be a clear image corresponding to the same target object.

[0115] The calibration images can be obtained through at least one of the following methods: scientific imaging equipment, digital camera, calibration library, and super-resolution dataset.

[0116] The scientific imaging device may have a full-frame image sensor with a resolution of no less than 60 million pixels; an analog-to-digital conversion bit depth of no less than 14 bits; a dynamic range of no less than 12 stops (e.g., based on ISO 100); and be equipped with a global shutter and an active cooling module (e.g., operating temperature fluctuation within ±1°C).

[0117] The digital camera may have a resolution between 20 million and 50 million pixels; its dynamic range may not exceed 11 stops (e.g., based on ISO 100); it may employ a rolling shutter mechanism; and it may not integrate radiometric calibration.

[0118] The calibration library may be the MIT-Adobe FiveK Dataset, which includes 5,000 RAW format images.

[0119] The super-resolution dataset can be DIV2K, which has a 2K resolution and contains 800 training images and 100 validation images.

[0120] In one embodiment, the configuration method may further include the step of: performing color space conversion on the calibration image. This step may be set before step S1. Specifically, it may include uniformly converting the calibration image to the prior RGB color space (e.g., sRGB→CIE XYZ conversion).

[0121] In one embodiment, the configuration method may further include the step of normalizing pixel values ​​in the calibration image. This step may be set before step S1. Specifically, it may include normalizing the pixel values ​​in the calibration image to the range of [0, 1]; for example, it may be implemented based on the Camera Response Function (CRF).

[0122] In one embodiment, the configuration method may further include the step of randomly cropping the calibration image. This step may be set before step S1. Specifically, it may include randomly cropping the calibration image into 512 pixel × 512 pixel sub-image blocks, thereby ensuring that the overlap rate between the sub-image blocks is less than or equal to 20%.

[0123] The loss value characterizes the difference between the aberration-corrected image and the calibration image.

[0124] The loss value can be determined according to a preset loss function. The loss function can be a pixel-level loss function, a frequency domain loss function, a perceptual loss function, or a multi-scale loss function. Specifically, the loss function can be one of the following: mean squared error loss, mean absolute error loss, or pyramid loss.

[0125] The model configuration is updated based on the loss value. Specifically, the calibration image can be used as the optimization target to update the model configuration so that the newly generated aberration-corrected image is similar to the calibration image.

[0126] The optical imaging system thus determined can correct aberrations in the imaging information through frequency domain processing of the first model and spatial domain processing of the second model. Even if the configuration of the optical module introduces multimodal aberrations, the configuration method provided in this application can be used to correct them in a targeted manner.

[0127] In one scenario, when the optical module is changed, the configuration method provided in this application can be executed to quickly obtain a first model and a second model adapted to the new optical module, which have adaptive aberration correction capabilities.

[0128] In another scenario, if the business requires the optical imaging system to adapt to harsh conditions with more modal aberrations, then the corresponding aberrations can be introduced into the modulated optical signal, imaging information, first correction information, or aberration correction image according to the aberration correction task. By executing the aforementioned steps S1 to S3, the system can adapt to the specific aberration correction requirements.

[0129] In one embodiment, the configuration method may further include the step of updating the configuration of the optical module based on the loss value.

[0130] An optical module may include several optical elements. The configuration of the optical module may include the combination of optical elements and the spacing between the optical elements.

[0131] The optical module may include refractive optical elements. The configuration of the optical module may include refractive index and radius of curvature.

[0132] The optical module may include diffractive optical elements. The configuration of the optical module may include focal length characteristics, phase function of the diffraction surface, radial radius at the abrupt change of each zone on the diffraction surface, zone depth of the diffraction surface, and diffraction efficiency.

[0133] Furthermore, concepts or definitions with the same expressions and configurations of components that point to the same functions among the various implementation methods and embodiments of this application can be referenced and appropriated from each other, and will not be repeated hereafter.

[0134] In one embodiment, the optical module includes metasurface optical elements.

[0135] Existing optical imaging systems incorporating metasurface optical elements typically suffer from significant multimodal aberrations during imaging, including chromatic aberration (axial / lateral), spherical aberration, asymmetric astigmatism, field curvature distortion, and geometric distortion. However, in this application, through the configuration method described above, the first and second models can adapt to the correction of multimodal aberrations, thereby obtaining a clear image of the target object. In other words, the optical imaging system configured according to the method of this application has a more prominent advantage in scenarios utilizing metasurface optical elements for imaging.

[0136] Metasurface optical elements can be configured based on any of the technical solutions in this application.

[0137] In one embodiment, the configuration method may further include the step of updating the configuration of the metasurface optical element based on the loss value.

[0138] The configuration of the metasurface optical element may include the arrangement period, material, shape, size, and position coordinates of the nanostructures therein.

[0139] In one embodiment, a first model is used to generate a frequency domain representation based on imaging information. The first model is also used to denoise the frequency domain representation based on a first set of parameters to determine first correction information.

[0140] In one embodiment, the configuration method includes the step of: updating the first parameter group based on the loss value to update the configuration of the first model.

[0141] This step can be included in step S3, and in particular in the step of "updating the configuration of the first model based on the loss value".

[0142] In this way, frequency domain aberrations and noise are suppressed in a targeted manner, preliminary correction is achieved, the burden on the second model is reduced, and the overall convergence of the model is accelerated.

[0143] In one embodiment, the first model generates a frequency domain representation based on the imaging information through Fourier transform.

[0144] The frequency domain representation of imaging information It can satisfy: .

[0145] These are spatial coordinates; They are continuous frequency domain coordinates; It is imaging information; It is a transformation kernel.

[0146] In one embodiment, the first model generates a frequency domain representation based on the imaging information through discrete Fourier transform.

[0147] The frequency domain representation It can satisfy: .

[0148] It is the image size (in pixels) corresponding to the imaging information; These are the pixel coordinates in the spatial domain after the imaging information has been discretized. These are the frequency domain coordinates after the imaging information has been discretized.

[0149] The above parameters can be further satisfied: , . The spatial sampling interval in the x-direction. The spatial sampling interval is in the y-direction.

[0150] In one embodiment, the first model performs linear filtering denoising on the frequency domain representation based on gain parameters and bias parameters to determine first correction information.

[0151] In one embodiment, the configuration method includes the steps of: updating the gain parameter, updating the bias parameter, or updating both the gain parameter and the bias parameter based on the loss value, to update the configuration of the first model.

[0152] This step can be included in step S3, and in particular in the step of "updating the configuration of the first model based on the loss value".

[0153] Thus, by leveraging the advantages of linear filtering denoising—high computational efficiency, ease of implementation, stable training, and low risk of overfitting—we can provide a guarantee for the subsequent processing of the second model in the spatial domain.

[0154] In this embodiment, the first correction information It can satisfy: .

[0155] It is the gain parameter; It is the bias parameter.

[0156] In other embodiments, the linear filtering denoising can also be Wiener filtering denoising.

[0157] In one embodiment, the first model performs nonlinear filtering denoising on the frequency domain representation based on weight parameters to determine first correction information. The weight parameters may be model parameters of the first model.

[0158] In one embodiment, the configuration method includes the step of: updating at least one of several weight parameters based on the loss value to update the configuration of the first model.

[0159] This step can be included in step S3, and in particular in the step of "updating the configuration of the first model based on the loss value".

[0160] Thus, by utilizing the powerful nonlinear mapping capability of nonlinear filtering denoising, targeted compensation and frequency adaptation can be achieved, resulting in stronger robustness.

[0161] The nonlinear filtering denoising can be nonlinear Wiener filtering denoising, wavelet thresholding denoising, or sparse representation and dictionary learning denoising.

[0162] In one embodiment, the first model may sequentially perform linear filtering denoising and nonlinear filtering denoising on the frequency domain representation.

[0163] In one embodiment, the first correction information is determined based on the imaging information and the point spread function of the optical module.

[0164] In this embodiment, the first correction information It can satisfy: .

[0165] It is the frequency domain representation of the point spread function of the optical module.

[0166] In this way, it can accurately denoise and correct aberrations or other degradations caused by optical modules, more effectively restore image details, and improve the convergence speed and stability of the correction model. It is especially suitable for correcting complex aberrations caused by optical imaging systems.

[0167] In one embodiment, the first model performs nonlinear filtering denoising on the frequency domain representation based on weight parameters and the point spread function of the optical module to determine the first correction information.

[0168] The nonlinear filtering denoising can be weighted adaptive Wiener filtering denoising.

[0169] In one embodiment, the frequency domain representation of the first correction information satisfy: .

[0170] It is the frequency domain representation of imaging information; This is the frequency domain representation of the point spread function of the optical module. Each term on the right-hand side of the equation has a dimensionally similar to the frequency domain representation of the first correction information on the left-hand side. Alignment: In practical applications, all physical quantities are normalized during the preprocessing stage, so there is no need to explicitly consider the dimension issue.

[0171] In one specific embodiment, the range of values ​​for the trainable weight parameters is designed as follows: The value range is (0, 2).

[0172] Values ​​range from Between the minimum and maximum values.

[0173] The value range is (0, 1). In practical applications, It can be simplified to a frequency-independent trainable scalar, which still meets the performance requirements.

[0174] Values ​​range from Between the minimum and maximum values.

[0175] The value range is (0, 1).

[0176] Values ​​range from Between the minimum and maximum values.

[0177] The design of the dimensions of the trainable parameters above ensures that the dimensions are consistent when the terms in the formula are added together. The trainable weight parameters can be the model parameters of the first model.

[0178] In this embodiment, It can be considered as the zero-order part of filtering and regularization, which can be used to improve aberrations, noise, blurring effects or other degradations caused by the optical module itself. This is the overall weight parameter for this part; It can be used for regularization to improve training flexibility.

[0179] It can be considered as the frequency domain correlation enhancement part, which can be used to enhance or suppress specific frequency components. It is a weighting parameter for enhancing frequency domain correlation.

[0180] This can be considered as the bias component.

[0181] This can be considered as the frequency-selective attenuation component. Some of these are used for personalized attenuation control across different frequency bands; It can be used to control the cutoff frequency. The frequency domain representation of part of the imaging information The product can be spatial frequency coordinates The Adama product. This is the overall weight parameter for this part.

[0182] This embodiment extends traditional frequency domain filtering (such as Wiener filtering) into a trainable function approximation structure composed of multiple interpretable frequency domain operators. While maintaining the clarity of physical meaning, it significantly improves the robustness and adaptability of the system under complex noise conditions, and avoids the performance degradation caused by relying solely on a single statistical ratio model.

[0183] In one embodiment, the first model can also generate the corresponding spatial domain representation based on the denoised frequency domain representation through inverse Fourier transform to obtain the first correction information.

[0184] First calibration information It can satisfy: .

[0185] This represents the inverse Fourier transform.

[0186] Figure 3 A schematic diagram of imaging information is shown. Figure 4 A schematic diagram of the first correction information corresponding to the imaging information is shown. It can be seen that after frequency domain processing of the imaging information according to the first model, the resulting first correction information improves aspects such as aberrations, noise, blurring effects, or other degradations, and recovers image details to some extent.

[0187] In one embodiment, the second model is constructed based on a convolutional neural network.

[0188] In one embodiment, the configuration method includes the step of: updating the weights or biases of the convolutional neural network based on the loss value to update the configuration of the second model.

[0189] This step can be included in step S3, and in particular in the step of "updating the configuration of the second model based on the loss value".

[0190] In this embodiment, the convolutional neural network may include convolutional layers and normalization layers.

[0191] Based on the loss value, the weights or biases of the convolutional kernels in the convolutional layer can be updated to enhance the model's feature extraction capability; based on the loss value, the scaling parameters or offset parameters of the normalization layer can also be updated to accelerate convergence.

[0192] In one embodiment, the second model is constructed based on a generative adversarial neural network.

[0193] In one embodiment, the configuration method includes the step of: updating the weights or biases of the generative adversarial neural network based on the loss value to update the configuration of the second model.

[0194] This step can be included in step S3, and in particular in the step of "updating the configuration of the second model based on the loss value".

[0195] In this embodiment, the generative adversarial neural network may include a generator and a discriminator.

[0196] Based on the loss value, the weights or biases of the convolutional kernels in the generator can be updated; based on the loss value, the weights or biases of the convolutional kernels in the discriminator, or the parameters of the normalization layer therein, can also be updated.

[0197] In one embodiment, the aberration-corrected image is obtained by an optical imaging system imaging the target object. Specifically, an optical module receives light signals from the target object and generates modulated light signals; a sensing unit generates imaging information based on the modulated light signals; and the imaging information is processed by a first model and a second model to determine the aberration-corrected image.

[0198] In one embodiment, the aberration-corrected image is determined by a simulation model corresponding to the optical imaging system based on the target object. The optical module and sensing unit are abstracted into mathematical models. Information about the target object is input into these mathematical models to obtain ideal imaging information, which is then processed by a first model and a second model to determine the aberration-corrected image.

[0199] In one embodiment, the aberration-corrected image is calculated and determined by introducing the target aberration into the calibration image. Specifically, the aberration-corrected image is determined by calculating the introduction of the target aberration into the calibration image and then inputting the aberration-introduced calibration image into a first model and a second model.

[0200] In one embodiment, the aberration-corrected image is determined by calculating the calibration image based on the simulation model of the corresponding optical module and then by a second model. The optical module and sensing unit are abstracted as mathematical models. The calibration image is processed by these mathematical models to obtain imaging information simulating the imaging effect of the optical module. This information is then processed by the first and second models to determine the aberration-corrected image.

[0201] In one embodiment, the configuration method includes at least one of the following steps.

[0202] Step S01: Based on the point spread function of the optical module or the target aberration, the calibration image is degraded to generate degraded imaging information.

[0203] Step S02: Using the degraded imaging information as input to the first model, the aberration-corrected image is determined after frequency domain processing of the first model and spatial domain processing of the second model.

[0204] Steps S01 and S02 can be set before step S1, such as... Figure 5 As shown. Steps S01 and S02 can also be included in step S1, and in particular, they can be included in the step of "obtaining the aberration-corrected image".

[0205] In this way, the imaging process of the optical module is replaced and simulated by the point spread function or target aberration. The configuration process only requires the calibration image as input data, which not only facilitates the construction of a large amount of training data and eliminates the need for repeated imaging with the optical module, but also enables precise control or introduction of specific aberrations and can cover process deviations, thereby improving the robustness of the configured system.

[0206] The point spread function of the optical module is used to simulate the imaging effect of the optical module.

[0207] The target aberration can be determined based on the hardware configuration and application scenario of the optical imaging system, and is used to simulate the aberrations that may exist or need to be overcome during the implementation of the optical imaging system.

[0208] In one embodiment, the calibration image can be degraded simultaneously based on the point spread function of the optical module and the target aberration.

[0209] In one embodiment, the configuration method includes at least one of the following steps.

[0210] Step S011: Obtain the reference point diffusion function of the optical module.

[0211] Step S012: Introduce at least one of process tolerance disturbance or dynamic defocus simulation to the reference point diffusion function to determine the family of point diffusion functions.

[0212] Step S013: Perform convolution processing on the calibration image according to the family of point spread functions to generate degraded imaging information.

[0213] Steps S011 to S013 can be included in step S01.

[0214] The reference point diffusion function can be generated based on the theoretical optical model of the optical module; in one embodiment, the optical module includes metasurface optical elements, and the reference point diffusion function is generated based on the theoretical optical model composed of nanostructures in the metasurface.

[0215] The degraded imaging information generated in this embodiment is defined as the first degraded image. Then it can satisfy: .

[0216] It is a calibration image; It is a family of point spread functions; This represents the convolution operation.

[0217] The convolution process can be performed by randomly selecting a point spread function from the point spread function family and convolving it with the calibration image, or by convolving several point spread functions from the point spread function family with the calibration image respectively and taking the arithmetic mean or weighted average as the statistical result to determine the degraded imaging information.

[0218] Family of point spread functions It can be a set of point spread functions constructed based on the baseline point spread function. Point spread function family It can satisfy: .

[0219] These are the process parameters of the optical module, including but not limited to nanopillar size error, tilt angle, etc. This represents the statistical distribution corresponding to the process parameters.

[0220] The process tolerance disturbance can be a random deviation between the produced optical module and the ideal design caused by the physical limitations of the processing technology during the manufacturing process.

[0221] The process tolerance disturbances may include nanostructure size deviations, edge roughness (which can be simulated using Monte Carlo simulation), material property deviations, periodic alignment mismatches, and interlayer alignment errors.

[0222] The dynamic defocus can be a phenomenon in which the optical imaging system deviates from the ideal focus state due to changes in the relative position between the target object and the optical module and sensing unit or other environmental factors during the use of the optical module.

[0223] The dynamic defocus can be simulated by adjusting the defocus range through the wavefront aberration function, or by introducing a field curvature term into the wavefront aberration function to simulate field curvature defocus.

[0224] Thus, by applying degradation processing to the calibration image based on the family of point spread functions, the degradation process can cover the unexpected deviations between the actual manufacturing process and design parameters of the optical module, as well as the deviations that may occur during the use of the optical module, thereby improving the robustness of the system.

[0225] In this embodiment, the first degraded image can be directly used as the degraded imaging information. In other embodiments, the first degraded image can be subjected to noise addition and / or downsampling to determine the final degraded imaging information.

[0226] In one embodiment, the configuration method includes the step of: applying a noise component to the degraded image generated after degradation processing to generate degraded imaging information.

[0227] This step can be included in step S01. Specifically, this step can be included in step S013.

[0228] The degraded imaging information generated in this embodiment is defined as the second degraded image. Then it can satisfy: .

[0229] It is a noise component; It is a noise superposition operation.

[0230] In one embodiment, the noise component includes additive noise; in another embodiment, the noise component includes multiplicative noise; in yet another embodiment, the noise component includes both additive and multiplicative noise.

[0231] The additive noise can be Gaussian white noise, salt-and-pepper noise, or Poisson noise. The multiplicative noise can be speckle noise in coherent light imaging.

[0232] When the noise component includes additive noise Multiplicative noise At that time, the second degraded image It can satisfy: .

[0233] In this embodiment, the second degraded image can be directly used as the degraded imaging information.

[0234] In one embodiment, the configuration method includes the step of: performing downsampling processing on the degraded image generated after degradation processing based on a resolution reduction operator to generate degraded imaging information.

[0235] This step can be included in step S01. Specifically, this step can be included in step S013.

[0236] The degraded imaging information generated in this embodiment is defined as the third degraded image. Then it can satisfy: ;or .

[0237] It is a resolution reduction operator.

[0238] In one embodiment, the resolution reduction operator is used to perform pixel region averaging. In another embodiment, the resolution reduction operator is used to perform downsampling after anti-aliasing filtering. In yet another embodiment, the resolution reduction operator is used to perform spatial blurring based on convolution kernels. The above embodiments can be implemented independently, or in combination of two or three.

[0239] In one embodiment, the configuration method includes the steps of: applying a noise component to the degraded image generated after degradation processing, and performing downsampling processing on the degraded image with the applied noise component based on a resolution reduction operator to generate degraded imaging information.

[0240] This step can be included in step S01. Specifically, this step can be included in step S013.

[0241] The step of applying the noise component can be configured according to other embodiments of this application. The downsampling process can be configured according to other embodiments of this application.

[0242] Figure 6 A schematic diagram of a calibration image is shown. Figure 4 A schematic diagram of the aberration-corrected image corresponding to the calibration image is shown. It can be seen that the aberration-corrected image has undergone corresponding degradation processing compared to the calibration image.

[0243] Optical imaging methods One embodiment of this application provides an optical imaging method, such as... Figure 8 As shown.

[0244] The optical imaging method includes at least one of the following steps.

[0245] Step P1: Obtain imaging information.

[0246] The imaging information is generated based on a modulated optical signal. The modulated optical signal is generated by modulating incident light from the target object.

[0247] Step P2: The imaging information is processed in the frequency domain using the first model to generate the first correction information.

[0248] Step P3: Spatial processing of the first correction information is performed using the second model to generate an aberration-corrected image.

[0249] The configurations of the first model and the second model are determined based on the updated loss values. The loss values ​​are determined based on aberration-corrected images and calibration images corresponding to the same target object.

[0250] The first model can be configured based on any technical solution of this application. The configuration of the first model can be implemented according to any configuration method of this application. The second model can be configured based on any technical solution of this application. The configuration of the second model can be implemented according to any configuration method of this application.

[0251] Thus, since the loss value is constructed based on the aberration-corrected image and the calibration image, the first and second models trained accordingly have multimodal aberration recovery capabilities. By performing frequency domain correction and spatial domain correction on the imaging information in sequence through the two models, a clear image of the corresponding target object can be obtained.

[0252] In one embodiment, the optical imaging method includes at least one of the following steps.

[0253] Step P21: Based on the imaging information, the corresponding frequency domain representation is calculated using Fourier transform.

[0254] Step P22: Perform linear or nonlinear filtering denoising on the frequency domain representation to obtain the denoised frequency domain representation.

[0255] Step P23: Based on the denoised frequency domain representation, calculate the corresponding spatial domain representation through inverse Fourier transform to obtain the first correction information.

[0256] Steps P21 to P23 can be included in step P2.

[0257] The nonlinear filtering denoising can be weighted adaptive Wiener filtering denoising.

[0258] In one embodiment, the first correction information is determined based on the imaging information and the point spread function of the optical module.

[0259] In one embodiment, the optical imaging method may include the step of: performing nonlinear filtering denoising on the frequency domain representation using a first model, based on weight parameters and the point spread function of the optical module, to determine the first correction information.

[0260] In one embodiment, the frequency domain representation of the first correction information satisfy: .

[0261] It is the frequency domain representation of imaging information; It is the frequency domain representation of the point spread function of the optical module; , , , , , These are weight parameters.

[0262] In summary, the optical imaging system, configuration method, and optical imaging method provided in this application correct imaging information in the frequency and spatial domains by setting a first model and a second model respectively, thereby obtaining an aberration-corrected image at both global and local levels. This can cope with the combination of various complex optical variations corresponding to multimodal aberrations. Regarding the configuration of the optical imaging system, by using the aberration-corrected image and calibration image corresponding to the same target object, the configuration of the first model and / or the second model is adjusted in reverse with the calibration image as the optimization target, so that the final generated aberration-corrected image approximates the calibration image, further improving the accuracy of imaging in multimodal aberration scenarios and realizing hybrid optimization in the frequency and spatial domains. For different optical modules or different aberration environments, the model configuration most suitable for the current scenario can be determined, which has universal adaptability.

[0263] It should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0264] The detailed descriptions listed above are merely specific descriptions of feasible implementation methods of this application and are not intended to limit the scope of protection of this application. All equivalent implementation methods or modifications made without departing from the spirit of the art of this application should be included within the scope of protection of this application.

Claims

1. A method for configuring an optical imaging system, characterized in that, The optical imaging system includes: An optical module is used to modulate incident light from a target object, generating a modulated optical signal. The sensing unit is used to generate imaging information based on the modulated light signal. The first model is used to perform frequency domain processing on the imaging information to generate the first correction information. The second model is used to perform spatial processing on the first correction information to generate an aberration-corrected image. The configuration method includes: Obtain aberration-corrected image and calibration image corresponding to the same target object. Based on the aberration-corrected image and the calibration image, the loss value is determined. Based on the loss value, update the configuration of the first model, update the configuration of the second model, or update both the configuration of the first model and the configuration of the second model.

2. The configuration method according to claim 1, characterized in that, The optical module includes metasurface optical elements.

3. The configuration method according to claim 1, characterized in that, The first model is used to generate a frequency domain representation based on imaging information, and to perform denoising processing on the frequency domain representation based on a first set of parameters to determine the first correction information. The configuration method includes: Based on the loss value, the first parameter group is updated to update the configuration of the first model.

4. The configuration method according to claim 3, characterized in that, The first model performs linear filtering denoising on the frequency domain representation based on gain and bias parameters to determine the first correction information. The configuration method includes: Based on the loss value, update the gain parameter, update the bias parameter, or update both the gain parameter and the bias parameter to update the configuration of the first model.

5. The configuration method according to claim 3, characterized in that, The first model performs nonlinear filtering and denoising on the frequency domain representation based on weight parameters to determine the first correction information. The configuration method includes: Based on the loss value, at least one of several weight parameters is updated to update the configuration of the first model.

6. The configuration method according to claim 1 or 5, characterized in that, The first correction information is determined based on the imaging information and the dot spread function of the optical module; ， It is the frequency domain representation of the first correction information. It is the frequency domain representation of imaging information. It is the frequency domain representation of the point spread function of the optical module. , , , , , These are trainable weight parameters. It is a spatial frequency coordinate.

7. The configuration method according to claim 1, characterized in that, The second model is constructed based on a convolutional neural network or a generative adversarial neural network. The configuration method includes: Based on the loss value, the weights or biases of the neural network are updated to update the configuration of the second model.

8. The configuration method according to claim 1, characterized in that, Also includes: Based on the point spread function of the optical module or the target aberration, the calibration image is degraded to generate degraded imaging information. Using the degraded imaging information as input to the first model, the aberration-corrected image is determined through frequency domain processing of the first model and spatial domain processing of the second model.

9. The configuration method according to claim 8, characterized in that, include: Obtain the reference point spread function of the optical module. To determine the family of point spread functions, at least one of process tolerance perturbation or dynamic defocus simulation is introduced into the reference point spread function. The calibration image is convolved according to the family of point spread functions to generate the degraded imaging information.

10. The configuration method according to claim 9, characterized in that, Including one of the following: A noise component is applied to the degraded image generated after degradation processing to produce the degraded imaging information; the noise component includes additive noise, multiplicative noise, or both additive and multiplicative noise. The degraded image generated after degradation processing is downsampled based on a resolution reduction operator to generate the degraded imaging information; the resolution reduction operator is used to perform at least one of pixel region averaging, downsampling after anti-aliasing filtering, and spatial blurring based on convolution kernel; A noise component is applied to the degraded image generated after degradation processing, and the degraded image with the noise component applied is downsampled based on a resolution reduction operator to generate the degraded imaging information.

11. An optical imaging method, characterized in that, include: Imaging information is obtained, which is generated based on a modulated optical signal, which is generated by modulating incident light from the target object. The imaging information is processed in the frequency domain using the first model to generate first correction information. The first correction information is spatially processed by the second model to generate an aberration-corrected image. The configurations of the first model and the second model are determined based on the loss value update, which is based on the aberration-corrected image and the calibration image corresponding to the same target object.

12. The optical imaging method according to claim 11, characterized in that, include: Based on the imaging information, the corresponding frequency domain representation is calculated using Fourier transform. Perform linear or nonlinear filtering denoising on the frequency domain representation to obtain a denoised frequency domain representation. Based on the denoised frequency domain representation, the corresponding spatial domain representation is calculated through inverse Fourier transform to obtain the first correction information.

13. An optical imaging system, characterized in that, include: An optical module is used to modulate incident light from a target object, generating a modulated optical signal. The sensing unit is used to generate imaging information based on the modulated light signal. The first model is used to perform frequency domain processing on the imaging information to generate the first correction information. The second model is used to perform spatial processing on the first correction information to generate an aberration-corrected image. The configurations of the first model and the second model are determined based on the loss value update, which is based on the aberration-corrected image and the calibration image corresponding to the same target object.

14. The optical imaging system according to claim 13, characterized in that, The optical module includes metasurface optical elements.

15. The optical imaging system according to claim 13, characterized in that, The first model is used to denoise the imaging information. The second model is constructed based on a convolutional neural network or a generative adversarial neural network.

Images