Lensless dynamic imaging method and system based on neural field

Abstract

The invention discloses a lensless dynamic imaging method and system based on a neural field, and the method comprises the steps: obtaining multi-frame dynamic diffraction patterns at different heights through a multi-height lensless imaging device; training the initial neural network by using the multi-frame dynamic diffraction pattern to obtain a target neural network; obtaining a target moment corresponding to the input multi-frame dynamic diffraction pattern; and inputting the target moment into the target neural network to obtain a target reconstructed image corresponding to the target moment. Therefore, the time continuity between different frames is ensured, the information measured each time is utilized to the greatest extent, the high-resolution reconstruction of a dynamic object is realized, the time resolution and the image reconstruction accuracy are effectively improved, and the high-quality dynamic reconstruction is realized.

Description

Technical Field

[0001] The present invention relates to the field of microscopic imaging technology, and in particular to a lensless dynamic imaging method and system based on a neural field. Background Art

[0002] Currently, the amplitude and phase of light waves are the core physical quantities that describe their wave properties. Amplitude reflects an object's ability to reflect or transmit light, while phase contains information about the delay during light wave propagation and details of the object's internal structure. Therefore, capturing both the intensity (i.e., the square of the amplitude) and phase of light waves is essential for effective target detection.

[0003] Existing lensless imaging techniques can be used to obtain phase information of light waves. This technology reconstructs static objects by performing multiple measurements of a static light field. However, when the object is moving, traditional lensless imaging methods are no longer applicable. This problem can be alleviated by grouping the data from these multiple measurements into several groups and using the measurement results from each group to reconstruct the object at different moments in time. However, this method reduces the number of light field acquisitions at different moments, which affects the quality of the reconstruction at each moment and reduces the accuracy of the image reconstruction. Summary of the Invention

[0004] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0005] To this end, the present invention proposes a lensless dynamic imaging method based on neural fields, which can use multi-frame dynamic diffraction patterns to train the initial neural network to obtain the target neural network, and obtain the target reconstructed image corresponding to the target moment through the target neural network, thereby ensuring the temporal continuity between different frames, maximizing the use of information from each measurement, and achieving high-resolution reconstruction of dynamic objects, effectively improving the temporal resolution and the accuracy of image reconstruction, thereby achieving high-quality dynamic reconstruction.

[0006] Another object of the present invention is to propose a lensless dynamic imaging system based on neural fields.

[0007] To achieve the above objectives, the present invention provides a lensless dynamic imaging method based on a neural field, the method comprising:

[0008] Through a multi-height lensless imaging device, multiple frames of dynamic diffraction patterns at different heights are obtained;

[0009] Using the multi-frame dynamic diffraction patterns to train the initial neural network to obtain a target neural network;

[0010] Obtaining a target time corresponding to the input multiple frames of dynamic diffraction patterns;

[0011] input the target moment into the target neural network to obtain a target reconstruction image corresponding to the target moment.

[0012] The lens-free dynamic imaging method based on a neural field can further have the following additional technical features:

[0013] In an embodiment of the present application, the training of the initial neural network by using the plurality of dynamic diffraction images to obtain a target neural network comprises:

[0014] image reconstruction of the plurality of dynamic diffraction images by the initial neural network to obtain a predicted reconstruction image at each moment;

[0015] modeling of the predicted reconstruction image from an object plane to a sensor plane to obtain a contrast image of the predicted reconstruction image on the sensor plane;

[0016] loss value calculation of the contrast image and the plurality of dynamic diffraction images according to a loss function;

[0017] updating of network parameters in the initial neural network by using the loss value until the initial neural network converges, or, when the number of iterations of the network parameters reaches a preset number of times, obtaining a target neural network.

[0018] In an embodiment of the present application, the initial neural network comprises a first modeling module, a second modeling module, a data processing module and an output module; and the image reconstruction of the plurality of dynamic diffraction images by the initial neural network to obtain a predicted reconstruction image at each moment comprises:

[0019] modeling of a first feature image in the plurality of dynamic diffraction images by the first modeling module to obtain a corresponding image latent variable;

[0020] optical flow neural field modeling of each frame of dynamic image in the plurality of dynamic diffraction images by the second modeling module to obtain an optical flow value at each moment;

[0021] warping and MLP network processing of the image latent variable and the optical flow value by the data processing module to obtain a predicted reconstruction image at each moment;

[0022] output of the predicted reconstruction image at each moment by the output module.

[0023] In one embodiment of the present invention, modeling the predicted reconstructed image from the object plane to the sensor plane to obtain a comparison image of the predicted reconstructed image on the sensor plane includes: modeling the predicted reconstructed image from the object plane to the sensor plane using a conversion formula to obtain a comparison image of the predicted reconstructed image on the sensor plane, wherein the conversion formula is:

[0024]

[0025] wherein said |·| 2 Indicates taking the modulus value of the element and squaring it. is a comparison image of the predicted reconstructed image on the sensor plane, the OS t is the wavefront function of the sensor plane, λ is the wavelength, and stated Represent discrete Fourier transform and inverse discrete Fourier transform respectively, the H(f x ,f y ,z t ) is the propagation function, and the (f x ,f y ) is the spatial frequency, the O t To predict the reconstructed image, the z t It represents the distance between the sample plane and the sensor plane at time t.

[0026] In one embodiment of the present invention, the method further includes:

[0027] Obtaining a target reconstructed image corresponding to each moment of the multi-frame dynamic diffraction pattern through the target neural network;

[0028] The target reconstructed images are integrated in time sequence to obtain a corresponding dynamic video sequence.

[0029] To achieve the above objectives, the present invention further provides a lensless dynamic imaging system based on a neural field, the system comprising:

[0030] A first acquisition unit is configured to acquire multiple frames of dynamic diffraction patterns at different heights using a multi-height lensless imaging device;

[0031] A training unit, configured to train an initial neural network using the multiple frames of dynamic diffraction patterns to obtain a target neural network;

[0032] A second acquisition unit is used to input a target time corresponding to the multiple frames of dynamic diffraction patterns;

[0033] The reconstruction unit is used to input the target moment into the target neural network to obtain a target reconstructed image corresponding to the target moment.

[0034] The method comprises the following steps: acquiring multiple frames of dynamic diffraction images at different heights through a multi-height lensless imaging device; training an initial neural network by using the multiple frames of dynamic diffraction images to obtain a target neural network; acquiring a target time corresponding to the input multiple frames of dynamic diffraction images; inputting the target time into the target neural network to obtain a target reconstruction image corresponding to the target time. Thus, the initial neural network can be trained by using the multiple frames of dynamic diffraction images to obtain the target neural network, and the target reconstruction image corresponding to the target time can be obtained through the target neural network, so that the time continuity between different frames can be ensured, the information of each measurement can be maximized, the high-resolution reconstruction of a dynamic object can be realized, the time resolution and the accuracy of image reconstruction are effectively improved, and high-quality dynamic reconstruction is realized.

[0035] Additional aspects and advantages of the present application will be made apparent by the following description and the appended claims. BRIEF DESCRIPTION OF DRAWINGS

[0036] The above and / or additional aspects and advantages of the present application will become apparent and be more readily understood through consideration of the following description, taken in conjunction with the accompanying drawings, in which:

[0037] Figure 1 is a flow chart of a lensless dynamic imaging method based on a neural field according to an embodiment of the present application;

[0038] Figure 2 is a schematic diagram of a multi-height lensless imaging device according to an embodiment of the present application;

[0039] Figure 3 is a structural diagram of a lensless dynamic imaging system based on a neural field according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0041] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings and in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative labor should belong to the protection scope of the present application.

[0042] A neural field based lensless dynamic imaging method and system are described below with reference to the accompanying drawings according to an embodiment of the present application.

[0043] Figure 1 is a flowchart of a neural field based lensless dynamic imaging method according to an embodiment of the present application.

[0044] As shown in Figure 1 , the method comprises:

[0045] S1, acquiring multiple frames of dynamic diffraction images at different heights by a multi-height lensless imaging device;

[0046] In an embodiment of the present application, Figure 2 is a schematic diagram of a multi-height lensless imaging device according to the present application. As shown in Figure 2 , the top of the multi-height lensless imaging device provides illumination for a single-mode laser light source, and a plano-convex lens is used below for collimation to provide high coherence and uniform illumination conditions.

[0047] In an embodiment of the present application, by using the above multi-height lensless imaging device, an object can be fixed on a z-axis translation stage below, the sensor is below the translation stage, and after fixing the object, a blind reconstruction algorithm is used to measure the distance between the sensor and the object plane as the first frame distance z0, and then the translation stage is moved to different heights z t and the measurement values I t at different heights are recorded, so as to acquire multiple frames of dynamic diffraction images at different heights.

[0048] S2, training an initial neural network using the multiple frames of dynamic diffraction images to obtain a target neural network;

[0049] In an embodiment of the present application, after determining the multiple frames of dynamic diffraction images by the above steps, the initial neural network can be trained using the multiple frames of dynamic diffraction images to obtain the target neural network.

[0050] Specifically, in an embodiment of the present application, the method of training the initial neural network using the multiple frames of dynamic diffraction images to obtain the target neural network can comprise the following steps:

[0051] S21, performing image reconstruction on the multiple frames of dynamic diffraction images by the initial neural network to obtain predicted reconstruction images at each time;

[0052] S22, modeling the predicted reconstruction images from the object plane to the sensor plane to obtain a contrast image of the predicted reconstruction images on the sensor plane;

[0053] S23, calculating a loss value corresponding to the contrast image and the multiple frames of dynamic diffraction images according to a loss function;

[0054] S24, updating the network parameters in the initial neural network by using the loss value until the initial neural network converges, or obtaining a target neural network when the number of iterations of the network parameters reaches a preset number.

[0055] In an embodiment of the present application, the initial neural network can include a first modeling module, a second modeling module, a data processing module, and an output module.

[0056] In an embodiment of the present application, the method of image reconstruction of the multi-frame dynamic diffraction pattern by the initial neural network to obtain the predicted reconstruction image at each time can include the following steps:

[0057] S211, modeling the first frame feature pattern in the multi-frame dynamic diffraction pattern by the first modeling module to obtain the corresponding image latent variable;

[0058] S212, modeling the optical flow neural field of each frame of dynamic image in the multi-frame dynamic diffraction pattern by the second modeling module to obtain the optical flow value at each time;

[0059] S213, obtaining the predicted reconstruction image at each time by warping and MLP network processing of the image latent variable and the optical flow value by the data processing module;

[0060] S214, outputting the predicted reconstruction image at each time by the output module.

[0061] In an embodiment of the present application, the first modeling module models the first frame feature pattern in the multi-frame dynamic diffraction pattern as an implicit network parameter, that is, it can be represented by a learnable continuous function, which can generate the corresponding image latent variable according to the input spatial coordinates. For example, the learnable continuous function is f(x, y; \theta), where f is any neural network, (x, y) is the spatial coordinates, and \theta is the network parameter.

[0062] In an embodiment of the present application, the image latent variable can be a tensor of R C×H×W , where C, H, and W represent the channel, height, and width of the latent variable, respectively, and H and W are consistent with the size of the measurement data.

[0063] Further, in an embodiment of the present application, the second modeling module can model the optical flow neural field of each frame of dynamic image in the multi-frame dynamic diffraction pattern to obtain the optical flow value at each time, so as to capture and express the motion change of the object in time and space, thereby providing motion information for subsequent motion frame reconstruction and solving the problem caused by lack of time continuity in the prior art.

[0064] Further, in an embodiment of the present application, the predicted reconstruction image at each time can be obtained by warping and MLP network processing of the image latent variable and the optical flow value through the data processing module, wherein the warping and MLP network are the same as in the prior art, and specific descriptions can be referred to in the prior art, and the present disclosure embodiment will not be repeated here.

[0065] Further, in an embodiment of the present application, the mathematical process corresponding to the above processing process can be represented as:

[0066] of t (x,y)=f m (x,y,t)

[0067] O t =f o (warp(fe,of t ))

[0068] wherein f m (·) is the optical flow neural field, of t is the optical flow value at time t, fe is the first frame feature map, warp(·) is the warping operation, f o (·) is the refinement network, O t is the predicted reconstruction image of the object plane at time t, x and y represent two-dimensional spatial coordinates, and t represents time t.

[0069] In an embodiment of the present application, the method for modeling the predicted reconstruction image from the object plane to the sensor platform to obtain the corresponding image of the predicted reconstruction image on the sensor plane can comprise: modeling the predicted reconstruction image from the object plane to the sensor plane by a conversion formula to obtain the corresponding image of the predicted reconstruction image on the sensor plane, wherein the conversion formula is:

[0070]

[0071] wherein |·| 2 represents taking the modulus value and squaring element by element, is the corresponding image of the predicted reconstruction image on the sensor plane, OS t is the wavefront function of the sensor plane, and λ is the wavelength, and represent the discrete Fourier transform and the inverse discrete Fourier transform, respectively, H(f x ,f y ,z) is the propagation function, (f x ,f y ,z t ) is the spatial frequency, O t is the predicted reconstruction image, and z t represents the distance between the sample plane and the sensor plane at time t.

[0072] Further, in an embodiment of the present application, the reference image is obtained through the above steps and the multi-frame dynamic diffraction images I t Then, the loss value corresponding to the reference image and the multi-frame dynamic diffraction images can be calculated according to the loss function, and the network parameters in the initial neural network are updated using the loss value until the initial neural network converges, or when the number of iterations of the network parameters reaches a preset number, the target neural network is obtained. Wherein, by adjusting the network parameters of the initial neural network, the difference between the predicted measurement data and the actual measurement data can be minimized to obtain the best estimate of the first image and the optical flow field, so as to ensure that the target neural network can accurately represent the static and dynamic characteristics of the original object. In an embodiment of the present application, the loss function described above can be a loss function in the prior art, for example, L1Loss loss function.

[0073] In an embodiment of the present application, after the target neural network is obtained through the above steps, the reconstructed images corresponding to the multi-frame dynamic diffraction images can be automatically generated by the target neural network and stored.

[0074] S3, obtaining a target time corresponding to the input multi-frame dynamic diffraction images;

[0075] In an embodiment of the present application, the user can input a certain target time corresponding to the multi-frame dynamic diffraction images to obtain the reconstructed image corresponding to the target time. For example, assuming that the multi-frame dynamic diffraction images {I_10} include diffraction images corresponding to 10 time points, the target time corresponding to the input multi-frame dynamic diffraction images can be t1 (1 <= t1 <= 10).

[0076] S4, inputting the target time into the target neural network to obtain the target reconstructed image corresponding to the target time.

[0077] In an embodiment of the present application, after the target time is obtained through the above steps, the target time can be input into the target neural network to obtain the target reconstructed image corresponding to the target time.

[0078] In an embodiment of the present application, the above method can include: obtaining the target reconstructed images corresponding to each time point of the multi-frame dynamic diffraction images through the target neural network; and integrating the target reconstructed images in time sequence to obtain the corresponding dynamic video sequence, so that the continuity in time between different frames can be ensured, and the information of each measurement can be maximized to realize high-resolution reconstruction of dynamic objects. In addition, the modeling method of the neural field enables the above method to obtain the number of optical flows matching the number of measurements, effectively improving the time resolution.

[0079] According to an embodiment of the present application, a lens-free dynamic imaging method based on a neural field comprises: acquiring multiple frames of dynamic diffraction images at different heights by a multi-height lens-free imaging device; training an initial neural network using the multiple frames of dynamic diffraction images to obtain a target neural network; acquiring a target time corresponding to the input multiple frames of dynamic diffraction images; and inputting the target time into the target neural network to obtain a target reconstruction image corresponding to the target time. Thus, the initial neural network can be trained using the multiple frames of dynamic diffraction images to obtain the target neural network, and the target reconstruction image corresponding to the target time can be obtained through the target neural network, so that the time continuity between different frames can be ensured, the information of each measurement can be maximized, the high-resolution reconstruction of a dynamic object can be realized, the time resolution and the accuracy of image reconstruction are effectively improved, and high-quality dynamic reconstruction is realized.

[0080] To achieve the above-mentioned embodiments, as shown in the accompanying drawings, Figure 3 The present embodiment also provides a lens-free dynamic imaging system 10 based on a neural field, which comprises a first acquisition unit 301, a training unit 302, a second acquisition unit 303, and a reconstruction unit 304.

[0081] The first acquisition unit 301 is configured to acquire multiple frames of dynamic diffraction images at different heights by a multi-height lens-free imaging device.

[0082] The training unit 302 is configured to train an initial neural network using the multiple frames of dynamic diffraction images to obtain a target neural network.

[0083] The second acquisition unit 303 is configured to input a target time corresponding to the multiple frames of dynamic diffraction images.

[0084] The reconstruction unit 304 is configured to input the target time into the target neural network to obtain a target reconstruction image corresponding to the target time.

[0085] In an embodiment of the present application, the training unit 302 is specifically configured to:

[0086] perform image reconstruction on the multiple frames of dynamic diffraction images by the initial neural network to obtain predicted reconstruction images at different times;

[0087] model the predicted reconstruction images from an object plane to a sensor plane to obtain a contrast image of the predicted reconstruction images on the sensor plane;

[0088] calculate a loss value corresponding to the contrast image and the multiple frames of dynamic diffraction images according to a loss function;

[0089] update network parameters in the initial neural network using the loss value until the initial neural network converges, or the target neural network is obtained when the number of iterations of the network parameters reaches a preset number.

[0090] Further, the initial neural network comprises a first modeling module, a second modeling module, a data processing module and an output module; the training unit 302 is further configured to:

[0091] modeling the first frame of feature maps in the multi-frame dynamic diffraction image through the first modeling module to obtain corresponding image latent variables;

[0092] modeling the optical flow neural field of each frame of dynamic image in the multi-frame dynamic diffraction image through the second modeling module to obtain the optical flow value at each time;

[0093] obtaining the predicted reconstruction image at each time through the data processing module after warping and MLP network processing of the image latent variable and the optical flow value;

[0094] outputting the predicted reconstruction image at each time through the output module.

[0095] Further, the training unit 302 is further configured to:

[0096] modeling the predicted reconstruction image from the object plane to the sensor plane through the conversion formula to obtain the corresponding image of the predicted reconstruction image in the sensor plane, wherein the conversion formula is:

[0097]

[0098] wherein |·| 2 represents the element-wise modulo value and square, is the corresponding image of the predicted reconstruction image in the sensor plane, OS t is the wave front function of the sensor plane, and λ is the wavelength, and represent the discrete Fourier transform and the inverse discrete Fourier transform, respectively, H(f x ,f y ,z) is the propagation function, (f x ,f y ,z t ) is the spatial frequency, O t is the predicted reconstruction image, z t represents the distance between the sample plane and the sensor plane at time t.

[0099] Further, the system is further configured to:

[0100] obtaining the target reconstruction image corresponding to each time of the multi-frame dynamic diffraction image through the target neural network;

[0101] integrating the target reconstruction image in time sequence to obtain the corresponding dynamic video sequence.

[0102] According to the lens-free dynamic imaging system based on the neural field, the initial neural network can be trained by using multiple frames of dynamic diffraction patterns to obtain a target neural network, and a target reconstruction image corresponding to a target moment can be obtained through the target neural network, so that the continuity in time between different frames can be ensured, the information of each measurement can be maximized, the high-resolution reconstruction of a dynamic object can be realized, the time resolution and the accuracy of image reconstruction are effectively improved, and high-quality dynamic reconstruction is realized.

[0103] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, a person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0104] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise specifically limited.

Claims

1. A lensless dynamic imaging method based on neural field, characterized in that: include: Through a multi-height lensless imaging device, multiple frames of dynamic diffraction patterns at different heights are obtained; Using the multi-frame dynamic diffraction patterns to train the initial neural network to obtain a target neural network; Obtaining a target time corresponding to the input multiple frames of dynamic diffraction patterns; The target moment is input into the target neural network to obtain a target reconstructed image corresponding to the target moment.

2. The method according to claim 1, characterized in that The method of training the initial neural network using the multiple frames of dynamic diffraction patterns to obtain the target neural network includes: Reconstructing the multiple frames of dynamic diffraction patterns using the initial neural network to obtain predicted reconstructed images at each moment; Modeling the predicted reconstructed image from the object plane to the sensor plane to obtain a comparison image of the predicted reconstructed image on the sensor plane; Calculating loss values ​​corresponding to the control image and the multi-frame dynamic diffraction patterns according to a loss function; The loss value is used to update the network parameters in the initial neural network until the initial neural network converges, or when the number of iterations of the network parameters reaches a preset number, a target neural network is obtained.

3. The method according to claim 2, characterized in that The initial neural network includes a first modeling module, a second modeling module, a data processing module, and an output module; the image reconstruction of the multiple frames of dynamic diffraction patterns is performed by the initial neural network to obtain a predicted reconstructed image at each moment, including: Modeling the first frame feature map in the multi-frame dynamic diffraction map by the first modeling module to obtain corresponding image latent variables; Performing optical flow neural field modeling on each frame of the dynamic image in the multi-frame dynamic diffraction image by the second modeling module to obtain the optical flow value at each moment; After warping and MLP network processing are performed on the image latent variables and the optical flow values ​​by the data processing module, a predicted reconstructed image at each moment is obtained; The predicted reconstructed image at each moment is outputted through the output module.

4. The method according to claim 2, characterized in that The step of modeling the predicted reconstructed image from the object plane to the sensor plane to obtain a comparison image of the predicted reconstructed image on the sensor plane includes: modeling the predicted reconstructed image from the object plane to the sensor plane using a conversion formula to obtain a comparison image of the predicted reconstructed image on the sensor plane, wherein the conversion formula is: wherein said |·| 2 Indicates taking the modulus value of the element and squaring it. is a comparison image of the predicted reconstructed image on the sensor plane, the OS t is the wavefront function of the sensor plane, λ is the wavelength, and stated Represent discrete Fourier transform and inverse discrete Fourier transform respectively, the H(f x ,f y ,z t ) is the propagation function, and the (f x ,f y ) is the spatial frequency, the O t To predict the reconstructed image, the z t It represents the distance between the sample plane and the sensor plane at time t.

5. The method according to claim 1, wherein The method further comprises: Obtaining a target reconstructed image corresponding to each moment of the multi-frame dynamic diffraction pattern through the target neural network; The target reconstructed images are integrated in time sequence to obtain a corresponding dynamic video sequence.

6. A lensless dynamic imaging system based on neural field, characterized in that: include: A first acquisition unit is configured to acquire multiple frames of dynamic diffraction patterns at different heights using a multi-height lensless imaging device; A training unit, configured to train an initial neural network using the multiple frames of dynamic diffraction patterns to obtain a target neural network; A second acquisition unit is used to input a target time corresponding to the multiple frames of dynamic diffraction patterns; The reconstruction unit is used to input the target moment into the target neural network to obtain a target reconstructed image corresponding to the target moment.

7. The system according to claim 6, characterized in that The training module is specifically used to: Reconstructing the multiple frames of dynamic diffraction patterns using the initial neural network to obtain predicted reconstructed images at each moment; Modeling the predicted reconstructed image from the object plane to the sensor plane to obtain a comparison image of the predicted reconstructed image on the sensor plane; Calculating loss values ​​corresponding to the control image and the multi-frame dynamic diffraction patterns according to a loss function; The loss value is used to update the network parameters in the initial neural network until the initial neural network converges, or when the number of iterations of the network parameters reaches a preset number, a target neural network is obtained.

8. The system according to claim 7, characterized in that The initial neural network includes a first modeling module, a second modeling module, a data processing module and an output module; the training module is further used to: Modeling the first frame feature map in the multi-frame dynamic diffraction map by the first modeling module to obtain corresponding image latent variables; Performing optical flow neural field modeling on each frame of the dynamic image in the multi-frame dynamic diffraction image by the second modeling module to obtain the optical flow value at each moment; After warping and MLP network processing are performed on the image latent variables and the optical flow values ​​by the data processing module, a predicted reconstructed image at each moment is obtained; The predicted reconstructed image at each moment is outputted through the output module.

9. An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images