CN118154411A - Digital Adaptive Optics Architecture and System - Google Patents

Abstract

The present invention discloses a digital adaptive optical architecture and system, including rearranging pixels of an original light field image to obtain a sub-aperture image with a low spatial sampling rate; processing the sub-aperture image with a low spatial sampling rate using a TIS algorithm to obtain a sub-aperture image with a high spatial sampling rate; estimating the spatial offset of the sub-aperture image with a high spatial sampling rate using a wavefront gradient estimation algorithm from coarse granularity to fine granularity to obtain a first wavefront gradient, and converting the first wavefront gradient into a first wavefront aberration using a trained MLP model; performing a phase space deconvolution operation on the sub-aperture image with a high spatial sampling rate based on the first wavefront aberration to obtain an image reconstruction result of a local image. The present invention can perform high-speed wide-field wavefront detection, use a turbulence-induced scanning algorithm to increase the sampling rate, and use an incoherent aperture synthesis algorithm to achieve aberration removal and high-resolution imaging.

Description

Digital Adaptive Optics Architecture and System

Technical Field

The present invention relates to the field of light field imaging technology, and in particular to a digital adaptive optics architecture and system.

Background technique

Optical aberrations are common in natural environments and imaging systems. Atmospheric turbulence and imperfect lenses can cause light trajectories to bend, resulting in optical aberrations, blurry images and distorted signals. Adaptive optics (AO) technology aims to measure and correct optical aberrations caused by atmospheric turbulence or other media in real time. This technology is mainly used in astronomical observations, laser communications, visual science and other fields, and can significantly improve image quality and signal stability.

The core components of the AO system include wavefront sensors, deformable mirrors, and advanced control systems. The wavefront sensor is used to detect the distortion of the light wavefront entering the system in real time; the deformable mirror adjusts its shape according to the measurement results of the wavefront sensor to compensate for the distortion of the light wavefront; the control system coordinates the work of the wavefront sensor and the deformable mirror to ensure that the distortion is effectively corrected.

Existing AO systems have certain limitations in terms of field of view and system complexity. Optical aberrations have spatial inconsistency. Existing AO systems mainly use Shack-Hartmann wavefront sensors, which can only measure single optical aberrations within a small field of view (FOV) and cannot solve the problem of spatial inconsistency. Advanced AO technologies, such as MCAO and GLAO, use multi-sensor solutions to expand the field of view to several arc minutes, but sacrifice a certain degree of wavefront detection accuracy, resulting in increased system complexity.

Summary of the invention

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

To this end, the present invention proposes a digital adaptive image reconstruction method to solve the problem of wide field of view, high-resolution imaging and achieve high-resolution imaging when optical aberrations are large.

Another object of the present invention is to provide a digital adaptive image reconstruction system.

To achieve the above object, the present invention provides a digital adaptive image reconstruction method, comprising:

Rearrange the pixels of the original light field image to obtain a sub-aperture image with a low spatial sampling rate;

Processing the sub-aperture image with low spatial sampling rate by using TIS algorithm to obtain a sub-aperture image with high spatial sampling rate;

Using a wavefront gradient estimation algorithm from coarse granularity to fine granularity to estimate the spatial offset of the sub-aperture image with a high spatial sampling rate to obtain a first wavefront gradient, and using a trained MLP model to convert the first wavefront gradient into a first wavefront aberration;

A phase space deconvolution operation is performed on the sub-aperture image with a high spatial sampling rate based on the first wavefront aberration to obtain an image reconstruction result of a local image.

The digital adaptive image reconstruction method of the embodiment of the present invention may also have the following additional technical features:

In one embodiment of the present invention, after obtaining the sub-aperture image with a low spatial sampling rate, the method further includes:

Acquire an image sequence of sub-aperture images of the same sub-aperture at different times and with the low spatial sampling rate;

The relative coordinates between the sub-aperture images at the reference time are calculated according to the relative offset between the image sequences.

In one embodiment of the present invention, the sub-aperture image with a low spatial sampling rate is processed by using a TIS algorithm to obtain a sub-aperture image with a high spatial sampling rate, including:

splicing the multiple sub-aperture images in the image sequence according to the relative coordinates of the sub-aperture images at a reference time using a TIS algorithm to obtain a coordinate splicing result;

A sub-aperture image with a high spatial sampling rate at a reference time is obtained based on the coordinate stitching result.

In one embodiment of the present invention, using a trained MLP model to convert the first wavefront gradient into a first wavefront aberration includes:

Obtaining a second wavefront gradient obtained by a coarse-grained to fine-grained wavefront gradient estimation algorithm;

The second wavefront gradient is converted into a second wavefront aberration by using a two-dimensional integration algorithm, so as to fit the Zernike polynomial coefficients to be fitted corresponding to the second wavefront aberration;

The MLP model is trained using the second wavefront gradient to obtain a trained MLP model; wherein the model input layer corresponds to the dimension of the second wavefront gradient, and the number of nodes in the model output layer is equal to the number of Zernike polynomial coefficients to be fitted;

The first wavefront gradient is input into the trained MLP model and the corresponding Zernike polynomial coefficients are output to calculate the first wavefront aberration.

In one embodiment of the present invention, after obtaining the image reconstruction result of the local image, the method further includes:

Reconstruct each local image to obtain image reconstruction results of all local images;

The image reconstruction results of all the local images are spliced to obtain an image reconstruction result of the global image.

To achieve the above object, the present invention provides a spatiotemporal angular fusion dynamic light field reconstruction system, comprising:

An image pixel rearrangement module, used for rearranging pixels of the original light field image to obtain a sub-aperture image with a low spatial sampling rate;

An aperture image processing module, used for processing the sub-aperture image with low spatial sampling rate by using TIS algorithm to obtain a sub-aperture image with high spatial sampling rate;

A wavefront aberration conversion module, used to estimate the spatial offset of the sub-aperture image with a high spatial sampling rate by using a wavefront gradient estimation algorithm from coarse granularity to fine granularity to obtain a first wavefront gradient, and convert the first wavefront gradient into a first wavefront aberration by using a trained MLP model;

A local image reconstruction module is used to perform a phase space deconvolution operation on the sub-aperture image with a high spatial sampling rate based on the first wavefront aberration to obtain an image reconstruction result of a local image.

The digital adaptive image reconstruction method and system of the embodiment of the present invention performs high-speed wide-field-of-view wavefront detection through a plug-and-play wide-field-of-view wavefront sensor, adopts a turbulence-induced scanning algorithm to improve the sampling rate, and uses an incoherent aperture synthesis algorithm to achieve low-cost, wide-field-of-view, and high-resolution imaging.

Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and/or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

1 is a flow chart of a digital adaptive image reconstruction method according to an embodiment of the present invention;

FIG2 is a framework diagram of a digital adaptive image reconstruction method according to an embodiment of the present invention;

3 is a flow chart of a wavefront gradient estimation algorithm from coarse granularity to fine granularity according to an embodiment of the present invention;

FIG4 is a flow chart of an ISA algorithm according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a digital adaptive image reconstruction system according to an embodiment of the present invention.

Detailed ways

It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

The digital adaptive image reconstruction method and system according to the embodiments of the present invention are described below with reference to the accompanying drawings.

FIG. 1 is a flow chart of a digital adaptive image reconstruction method according to an embodiment of the present invention.

As shown in FIG1 , the method includes but is not limited to the following steps:

S1, rearrange the pixels of the original light field image to obtain a sub-aperture image with a low spatial sampling rate;

S2, using the TIS algorithm to process the sub-aperture image with a low spatial sampling rate to obtain a sub-aperture image with a high spatial sampling rate;

S3, using a wavefront gradient estimation algorithm from coarse-grained to fine-grained to estimate the spatial offset of the sub-aperture image with a high spatial sampling rate to obtain a first wavefront gradient, and using the trained MLP model to convert the first wavefront gradient into a first wavefront aberration;

S4, performing a phase space deconvolution operation on the sub-aperture image with a high spatial sampling rate based on the first wavefront aberration to obtain an image reconstruction result of the local image.

It is understandable that the present invention proposes a new DAO architecture, the core component of which is a plug-and-play wide-field wavefront sensor (WWS). In terms of hardware, WWS is just a minor modification of the traditional 2D CMOS image sensor. Specifically, it is only necessary to place the microlens array (MLA) on the original image plane and then place the 2D CMOS image sensor on the back focal plane of the MLA. The system complexity and implementation cost are low. In terms of algorithms, the present invention adopts a wavefront gradient estimation algorithm from coarse-grained to fine-grained, combined with multi-layer perceptron (MLP) mapping, to achieve high-speed (30Hz), wide-field (over 1100 arc seconds) spatial non-uniform aberration detection.

FIG2 is a framework diagram of the digital adaptive image reconstruction method of the present invention, which includes a preprocessing stage, a turbulence induced scanning (TIS) algorithm flow, an aberration wavefront detection and an incoherent aperture synthesis (ISA) algorithm flow as shown in FIG2 .

In one embodiment of the present invention, the pre-processing stage is a pixel rearrangement operation.

Specifically, the original light field image is converted into several corresponding sub-aperture images, each of which represents the projection of the light field at a different angle. The slight deviation between CMOS and MLA can be corrected by resizing and rotating operations.

In one embodiment of the present invention, TIS, i.e., the left branch of FIG. 2 , implements sub-pixel level image alignment and stitching.

It is understandable that light field imaging uses 2D CMOS to record 4D light field information, sacrificing the spatial sampling rate while recording the angle information. Therefore, the spatial sampling rate of the sub-aperture image obtained by preprocessing is too low, and TIS is used to restore its sampling rate.

Specifically, the relative offset of the image sequence of the same sub-aperture at different times is first estimated to obtain the relative coordinates between the sub-aperture images; then, the multiple sub-aperture images in the sequence are spliced according to the relative coordinates of the sub-aperture images at the reference time by the TIS algorithm, so as to obtain the sub-aperture image at the reference time with a high spatial sampling rate. The image obtained by the TIS algorithm has high quality and no obvious dynamic artifacts.

The TIS algorithm of the embodiment of the present invention realizes sub-pixel image alignment and stitching. The algorithm input is multiple pictures with low spatial sampling rate, and the output is a picture with high spatial sampling rate. When the spatial sampling rate is low, the optical aberration caused by the rapidly changing atmospheric turbulence can be regarded as a fast scan of spatial inconsistency. The TIS algorithm can fuse the scanning information and improve the spatial sampling rate.

The TIS algorithm process can be expressed by the following formula:

In the formula, Indicates the number of image frames used for TIS (required to be an odd number, then the reference time/> ), Indicates time /> The corresponding /> sub-aperture images, /> represents the TIS result of the sub-aperture image at the reference time,/> represents the two-dimensional coordinates on the image plane,/> Indicates/> With/> The relative coordinate offset between Represents the sub-aperture image/> At coordinate position /> The result of using bicubic interpolation (BI),/> represents the mean square error, /> Represents scattered point interpolation (SI).

The purpose of formula (1) is to calculate the relative coordinate offset of the sub-aperture images at different times. The optimization problem is solved by the stochastic gradient descent algorithm (SGD). The meaning of formula (2) is to use the scattered point interpolation method to obtain the grayscale values corresponding to the uniform grid positions of multiple sub-aperture images with randomly distributed lateral coordinates and fuse them into a high-resolution sub-aperture image.

In one embodiment of the present invention, the aberration wavefront detection is the right branch of FIG. 2 .

Specifically, the wavefront gradient estimation algorithm from coarse granularity to fine granularity is used to estimate the spatial non-uniform offset of multiple sub-aperture images at a specific moment. The offset is the wavefront gradient field. The "coarse granularity to fine granularity" optimization strategy can significantly improve the robustness and running speed of the algorithm. The specific process of the algorithm is shown in Figure 3, and its process is briefly described below:

(1) This algorithm is used to estimate the target sub-aperture image To the reference sub-aperture image/> In the present invention, for different sub-aperture images captured at the same time, a reference sub-aperture image is selected. , estimate all other sub-aperture images/> To/> The spatially non-uniform lateral offsets are spliced together according to the position of the aperture plane to obtain the spatially non-uniform wavefront gradient field;

(2) First, refer to the process in the upper part of Figure 3. With/> Standardization is performed to remove its intensity distribution to avoid inconsistent intensity distribution causing failure of the estimation algorithm. After that, the sub-aperture image is normalized. With/> The wavefront gradient is estimated and the optimization problem is expressed as follows:

In the formula represents the mean square error, /> represents bicubic interpolation, /> represents bicubic upsampling,/> represents the two-dimensional coordinates on the image plane (same size as the image),/> Represents the two-dimensional coordinates on the image plane (the size is the same as the number of local spatial consistent regions, smaller than the image size), /> Indicates the lateral offset;

(3) Referring to the process in the lower part of Figure 3, first optimize the coarse-grained wavefront gradient , then /> The value of is used to initialize the fine-grained wavefront gradient/> , and further optimize/> , after the iterative process converges,/> That is what you want/> Both coarse-grained and fine-grained wavefront gradients are optimized using the stochastic gradient descent algorithm, and the optimizer is Adam.

Furthermore, after obtaining the wavefront gradient, the wavefront aberration can be obtained from the wavefront gradient using an integral algorithm. In practical application scenarios, the integral algorithm is slow and cannot meet the real-time observation requirements. For this reason, the present invention designs and trains an MLP model to achieve rapid conversion from gradient to aberration. The MLP is trained in a supervised mode, and the trained MLP is used to replace the integral process for reasoning. The specific process is as follows:

1) Prepare input data: The input data is the second wavefront gradient calculated by the coarse-grained to fine-grained wavefront gradient estimation algorithm;

2) Prepare the output to be fitted: Run a two-dimensional integration algorithm on the second wavefront gradient to obtain the second wavefront aberration, and then use the least squares method to fit the Zernike polynomial coefficients corresponding to the second wavefront aberration. These Zernike polynomial coefficients are the output to be fitted;

3) Initialization: Initialize the MLP network with a double hidden layer structure. The input layer is the wavefront gradient. The first hidden layer has 300 nodes, the second hidden layer has 500 nodes, and the output layer is the Zernike polynomial coefficients. The parameters are randomly initialized.

4) Training: The loss function uses MSE, the optimizer uses Adam, and the training is iterated until the MSE loss converges;

5) Reasoning: If the first wavefront gradient is input to the MLP, the MLP can output the Zernike polynomial coefficients of the first wavefront aberration.

Therefore, the experiment found that the MLP model can fit the training data well, and its performance on the validation data is very close to the training data, and it has good generalization ability.

In one embodiment of the present invention, the specific process of the ISA algorithm is shown in FIG4 , and the process is briefly described below:

Considering the spatial inconsistency of aberrations, the reconstruction is performed by block deconvolution, and the deconvolution method uses the phase space deconvolution algorithm;

For the local area, the aberration is considered to be consistent, so the obtained wavefront aberration is used to run the phase space deconvolution algorithm on the obtained local sub-aperture image after TIS to obtain the reconstruction result of the local image;

By reconstructing each local area and obtaining the reconstruction results of all local areas, the reconstruction result of the global image can be obtained by simply splicing them together, thus achieving an improvement in global resolution.

It can be known that the present invention adopts the ISA algorithm to remove spatially non-uniform optical aberrations, thereby significantly improving imaging resolution.

For example, deep neural networks may replace the current aberration wavefront detection algorithm, and its accuracy and speed need to be verified. Deep neural networks may replace the block deconvolution operation in ISA, and its imaging quality and speed need to be verified.

In summary, the beneficial effects and application scenarios brought by the present invention include:

1) Low cost, wide field of view, high resolution imaging: The main problem solved by the present invention is wide field of view, high resolution imaging, which is particularly effective in cases where optical aberrations are large. In addition, the low cost feature of the present invention is conducive to its wide application in astronomical observation, long-distance imaging and other scenarios, paving the way for scientific exploration.

2) Turbulence analysis: Wide-field wavefront gradient detection is beneficial for analyzing atmospheric turbulence motion. For example, in Slope Detection and Ranging (SLODAR), wide-field wavefront gradient data is beneficial for improving the longitudinal resolution of turbulence profiling and the robustness of analysis results.

3) Turbulence prediction: According to the Taylor frozen flow hypothesis, wide-field turbulence aberration data is conducive to improving the accuracy of turbulence aberration prediction. In free-space optical communication, pre-compensation of turbulence aberration is an important link to improve the signal-to-noise ratio and reduce the bit error rate. The present invention is conducive to improving the prediction accuracy of turbulence aberration, thereby improving the communication quality.

According to the digital adaptive image reconstruction method of the embodiment of the present invention, high-speed wide-field-of-view wavefront detection is performed through a plug-and-play wide-field-of-view wavefront sensor, and sub-pixel image alignment and stitching are achieved using the TIS algorithm. The algorithm input is a plurality of pictures with low spatial sampling rates, and the output is a picture with high spatial sampling rates. When the spatial sampling rate is low, the optical aberrations caused by the rapidly changing atmospheric turbulence can be regarded as rapid scans of spatial inconsistency. The TIS algorithm can fuse the scanning information and improve the spatial sampling rate. The ISA algorithm is used to remove spatially inconsistent optical aberrations, significantly improving the imaging resolution.

In order to implement the above embodiment, as shown in Figure 5, a digital adaptive image reconstruction system 10 is also provided in this embodiment. The system 10 includes an image pixel rearrangement module 100, an aperture image processing module 200, a wavefront aberration conversion module 300 and a local image reconstruction module 400.

An image pixel rearrangement module 100 is used to rearrange pixels of an original light field image to obtain a sub-aperture image with a low spatial sampling rate;

The aperture image processing module 200 is used to process the sub-aperture image with a low spatial sampling rate using the TIS algorithm to obtain a sub-aperture image with a high spatial sampling rate;

The wavefront aberration conversion module 300 is used to estimate the spatial offset of the sub-aperture image with a high spatial sampling rate by using a wavefront gradient estimation algorithm from coarse granularity to fine granularity to obtain a first wavefront gradient, and convert the first wavefront gradient into a first wavefront aberration by using a trained MLP model;

The local image reconstruction module 400 is used to perform a phase space deconvolution operation on the sub-aperture image with a high spatial sampling rate based on the first wavefront aberration to obtain an image reconstruction result of the local image.

Furthermore, after the image pixel rearrangement module 100, it further includes: a coordinate calculation module, which is used to:

Acquire an image sequence of sub-aperture images of the same sub-aperture at different times and with a low spatial sampling rate;

The relative coordinates between the sub-aperture images at the reference time are calculated according to the relative offset between the image sequences.

Furthermore, the aperture image processing module 200 is also used for:

Using the TIS algorithm, multiple sub-aperture images in the image sequence are stitched according to the relative coordinates of the sub-aperture images at the reference time to obtain a coordinate stitching result;

Based on the coordinate stitching results, a sub-aperture image with high spatial sampling rate at the reference time is obtained.

Furthermore, the wavefront aberration conversion module 300 is also used for:

Obtaining a second wavefront gradient obtained by a coarse-grained to fine-grained wavefront gradient estimation algorithm;

The second wavefront gradient is converted into a second wavefront aberration by using a two-dimensional integration algorithm, so as to fit the Zernike polynomial coefficients to be fitted corresponding to the second wavefront aberration;

The MLP model is trained using the second wavefront gradient to obtain a trained MLP model; wherein the model input layer corresponds to the dimension of the second wavefront gradient, and the number of nodes in the model output layer is equal to the number of Zernike polynomial coefficients to be fitted;

The first wavefront gradient is input into the trained MLP model and the corresponding Zernike polynomial coefficients are output to calculate the first wavefront aberration.

Furthermore, after the local image reconstruction module 400, it further includes: a global image reconstruction module, which is used to:

Reconstruct each local image to obtain image reconstruction results of all local images;

The image reconstruction results of all local images are stitched together to obtain the image reconstruction result of the global image.

According to the digital adaptive image reconstruction system of the embodiment of the present invention, high-speed wide-field-of-view wavefront detection is performed through a plug-and-play wide-field-of-view wavefront sensor, and sub-pixel image alignment and stitching are realized by using the TIS algorithm. The algorithm input is multiple pictures with low spatial sampling rate, and the output is a picture with high spatial sampling rate. When the spatial sampling rate is low, the optical aberration caused by the rapidly changing atmospheric turbulence can be regarded as a rapid scan of spatial inconsistency. The TIS algorithm can fuse the scanning information and improve the spatial sampling rate. The ISA algorithm is used to remove spatially inconsistent optical aberrations, which significantly improves the imaging resolution.

In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

Claims (10)

1. A digital adaptive image reconstruction method, comprising: Rearrange the pixels of the original light field image to obtain a sub-aperture image with a low spatial sampling rate; Processing the sub-aperture image with low spatial sampling rate by using TIS algorithm to obtain a sub-aperture image with high spatial sampling rate; Using a wavefront gradient estimation algorithm from coarse granularity to fine granularity to estimate the spatial offset of the sub-aperture image with a high spatial sampling rate to obtain a first wavefront gradient, and using a trained MLP model to convert the first wavefront gradient into a first wavefront aberration; A phase space deconvolution operation is performed on the sub-aperture image with a high spatial sampling rate based on the first wavefront aberration to obtain an image reconstruction result of a local image. 2. The method according to claim 1, characterized in that after obtaining the sub-aperture image with a low spatial sampling rate, the method further comprises: Acquire an image sequence of sub-aperture images of the same sub-aperture at different times and with the low spatial sampling rate; The relative coordinates between the sub-aperture images at the reference time are calculated according to the relative offset between the image sequences. 3. The method according to claim 2, characterized in that the sub-aperture image with a low spatial sampling rate is processed by using a TIS algorithm to obtain a sub-aperture image with a high spatial sampling rate, comprising: splicing the multiple sub-aperture images in the image sequence according to the relative coordinates of the sub-aperture images at a reference time using a TIS algorithm to obtain a coordinate splicing result; A sub-aperture image with a high spatial sampling rate at a reference time is obtained based on the coordinate stitching result. 4. The method according to claim 1, characterized in that the first wavefront gradient is converted into a first wavefront aberration using a trained MLP model, comprising: Obtaining a second wavefront gradient obtained by a coarse-grained to fine-grained wavefront gradient estimation algorithm; The second wavefront gradient is converted into a second wavefront aberration by using a two-dimensional integration algorithm, so as to fit the Zernike polynomial coefficients to be fitted corresponding to the second wavefront aberration; The MLP model is trained using the second wavefront gradient to obtain a trained MLP model; wherein the model input layer corresponds to the dimension of the second wavefront gradient, and the number of nodes in the model output layer is equal to the number of Zernike polynomial coefficients to be fitted; The first wavefront gradient is input into the trained MLP model and the corresponding Zernike polynomial coefficients are output to calculate the first wavefront aberration. 5. The method according to claim 1, characterized in that after obtaining the image reconstruction result of the local image, the method further comprises: Reconstruct each local image to obtain image reconstruction results of all local images; The image reconstruction results of all the local images are spliced to obtain an image reconstruction result of the global image. 6. A digital adaptive image reconstruction system, comprising: An image pixel rearrangement module, used for rearranging pixels of the original light field image to obtain a sub-aperture image with a low spatial sampling rate; An aperture image processing module, used for processing the sub-aperture image with low spatial sampling rate by using TIS algorithm to obtain a sub-aperture image with high spatial sampling rate; A wavefront aberration conversion module, used to estimate the spatial offset of the sub-aperture image with a high spatial sampling rate by using a wavefront gradient estimation algorithm from coarse granularity to fine granularity to obtain a first wavefront gradient, and convert the first wavefront gradient into a first wavefront aberration by using a trained MLP model; A local image reconstruction module is used to perform a phase space deconvolution operation on the sub-aperture image with a high spatial sampling rate based on the first wavefront aberration to obtain an image reconstruction result of a local image. 7. The system according to claim 6, characterized in that after the image pixel rearrangement module, it further comprises: a coordinate calculation module, for: Acquire an image sequence of sub-aperture images of the same sub-aperture at different times and with the low spatial sampling rate; The relative coordinates between the sub-aperture images at the reference time are calculated according to the relative offset between the image sequences. 8. The system according to claim 7, wherein the aperture image processing module is further used for: splicing the multiple sub-aperture images in the image sequence according to the relative coordinates of the sub-aperture images at a reference time using a TIS algorithm to obtain a coordinate splicing result; A sub-aperture image with a high spatial sampling rate at a reference time is obtained based on the coordinate stitching result. 9. The system according to claim 6, wherein the wavefront aberration conversion module is further used for: Obtaining a second wavefront gradient obtained by a coarse-grained to fine-grained wavefront gradient estimation algorithm; The second wavefront gradient is converted into a second wavefront aberration by using a two-dimensional integration algorithm, so as to fit the Zernike polynomial coefficients to be fitted corresponding to the second wavefront aberration; The MLP model is trained using the second wavefront gradient to obtain a trained MLP model; wherein the model input layer corresponds to the dimension of the second wavefront gradient, and the number of nodes in the model output layer is equal to the number of Zernike polynomial coefficients to be fitted; The first wavefront gradient is input into the trained MLP model and the corresponding Zernike polynomial coefficients are output to calculate the first wavefront aberration. 10. The system according to claim 6, characterized in that after the local image reconstruction module, it further comprises: a global image reconstruction module, which is used to: Reconstruct each local image to obtain image reconstruction results of all local images; The image reconstruction results of all the local images are spliced to obtain an image reconstruction result of the global image.