A cross-scale composite wavefront sensing method based on a deep learning model

Abstract

The application discloses a cross-scale composite wavefront sensing method based on a deep learning model, combines a Shack-Hartmann wavefront sensor (SHWFS) and off-axis digital holography (OADH) through a composite wavefront sensing system, fuses data of a wavefront measured by the composite wavefront sensing system by using an MSWM-Net model based on a deep neural network, realizes complementary advantages of performances of the two measurement systems, and further obtains a high-confidence reconstructed wavefront. When measuring a wavefront aberration containing different spatial frequencies, the method not only solves the problem of low phase accuracy in SHWFS measurement, but also solves the limitation of wrapping the wavefront in a period in OADH measurement. The feasibility and effectiveness of the method in wavefront detection are verified through tests. The application is more easy to realize quantitative measurement of atmospheric turbulence aberration, can provide design input parameters for an atmospheric turbulence aberration correction system, and can be used for distortion wavefront detection of the atmospheric turbulence correction system.

Description

Technical Field

[0001] This invention belongs to the field of adaptive optics and mainly relates to wavefront detection technology in adaptive optics systems, especially to cross-scale composite wavefront detection technology that combines the functions of Shaker-Hartmann wavefront detection system and off-axis digital holographic wavefront detection system. In particular, it relates to a cross-scale composite wavefront sensing method based on a deep learning model. Background Technology

[0002] Adaptive optics (AO) technology utilizes a control system to drive active devices to correct wavefront distortion, effectively improving the performance of optical systems and is widely used in astronomy, biomedicine, laser communication, and other fields. Wavefront sensors are a crucial component of adaptive optics systems. In wavefront measurement, the Shack-Hartmann Wavefront Sensor (SHWFS) is widely used due to its simple structure, high light energy utilization, and strong noise immunity. The Shack-Hartmann wavefront sensor mainly consists of a microlens array and a charge-coupled device (CCD). The basic principle is that the microlens array segments and samples the distorted wavefront, focusing the beam onto the CCD target surface. By calculating the centroid offset between the distorted wavefront and the planar wavefront, the average slope of the distorted wavefront within the sub-aperture range is further determined, and finally, a wavefront reconstruction algorithm is used to restore the wavefront. However, a trade-off must be struck between detection accuracy and dynamic range when designing the SHWFS. Off-axis digital holography (OADH) is an optical imaging technique that combines traditional optical holography with computer image processing. Compared to conventional imaging techniques, OADH offers advantages such as high precision, high spatiotemporal resolution, and high linearity. The basic principle involves the interference between the target probe wavefront and a reference beam; the interference fringes are focused onto the plane of a CCD camera, and the target probe wavefront is reconstructed using a holographic reconstruction algorithm.

[0003] When measuring wavefront aberrations at different scales (high and low frequencies), SHWFS can provide a large dynamic range, but its spatial resolution is limited due to its operating principle. OADH, on the other hand, can provide high spatial resolution, but its phase is affected by... The limitations of traditional single wavefront measurement devices mean that they cannot simultaneously meet the requirements of dynamic range and high spatial resolution wavefront measurement. Therefore, it is of great significance to conduct research on a multi-scale composite wavefront sensing method.

[0004] In recent years, deep learning technology has transformed input data nonlinearly into more abstract and higher-level representations, automatically learning corresponding features from massive datasets. This significantly reduces the need for modeling complex physical processes, giving it a strong advantage in processing high-dimensional data. Currently, deep learning has been extended into fields such as digital holographic imaging and SHWFS wavefront detection, including digital holography (from holograms to phase and intensity images of objects), phase unwrapping (from wrapped phase to absolute phase), and Shaker-Hartmann (reconstructing wavefront aberration phase or its Zernike coefficients). Furthermore, deep learning can automate and accelerate the analysis and processing of imaging data, thereby improving the efficiency and accuracy of data analysis. How to develop cross-scale composite wavefront sensing methods based on deep learning models is a problem that needs to be researched and solved. Summary of the Invention

[0005] The purpose of this invention is to address the challenge that existing traditional single-measurement systems cannot simultaneously meet the requirements of large dynamic range and high spatial resolution wavefront measurement. By analyzing the wavefront detection characteristics of SHWFS and OADH for practical physical systems, a composite wavefront sensing system is constructed. A deep learning-based MSWM-Net model is proposed to fuse the wavefront data detected by the composite wavefront sensing system, further obtaining a high-confidence reconstructed wavefront.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows: a cross-scale composite wavefront sensing method based on a deep learning model, wherein wavefront measurement is performed by a composite wavefront sensing system experimental device, which includes: a laser and an optical fiber beam splitter connected in sequence, and two branches connected in parallel after the optical fiber beam splitter. The first branch consists of a first collimating lens tube, an air outlet device, a fast reflector, a first beam splitter, a first beam shrinking lens tube, a microlens array, a first magnification matching system, and a first photodetector. The second branch consists of a second collimating lens tube, a second beam splitter, a second beam shrinking lens tube, a second magnification matching system, and a second photodetector. The first photodetector and the second photodetector are connected to a synchronous triggering device, which is used to realize the synchronous triggering of the laser and the first and second photodetectors.

[0007] The cross-scale composite wavefront sensing method based on deep learning models includes the following steps:

[0008] Step 1: First, the system is calibrated. This process includes: turning on the laser, the emitted beam passes through an optical fiber beam splitter to obtain two identical emitted beams, namely the first beam and the second beam. The first beam passes through a collimating lens tube, a fast reflector, and a first beam splitter before entering the first beam shrinking lens tube, and then exits onto a microlens array. After being split by the microlens array, the first beam passes through a first magnification matching system to form a calibration spot array image on the first photodetector, thus forming a calibration spot image. The second beam passes through a second collimating lens tube and then through a second beam splitter, interfering with the first beam that passed through the first beam splitter. After passing through the second beam shrinking lens tube, it is imaged on the second photodetector, forming an interference fringe image.

[0009] Step 2: Extract the calibration spot image from the first photodetector and the interference fringe image from the second photodetector;

[0010] Step 3: Turn on the air outlet and the laser. The emitted beam passes through the fiber optic beam splitter to obtain two identical emitted beams, namely the first beam and the second beam. The first beam is emitted through the first collimating lens tube to the air outlet. At this time, the first beam has aberrations. Then it passes through the fast reflector and the first beam splitter. The first beam splitter splits the first beam into a third beam and a fourth beam. The fourth beam passes through the second beam splitter, and the third beam enters the first beam shrinking lens tube and then exits to the microlens array. After being split by the microlens array, the third beam is imaged on the first photodetector in a dot matrix form with spatial offset relative to the calibration position through the first magnification matching system. The second beam passes through the second collimating lens tube and then the second beam splitter, interfering with the fourth beam. After passing through the second beam shrinking lens tube, it is imaged on the second photodetector to form an interference fringe image with aberrations.

[0011] Step 4: Extract the distorted spot image on the first photodetector, calculate the offset of the distorted spot relative to the calibration position using spot localization technology, calculate the wavefront slope using the offset, and then reconstruct the distorted wavefront using the wavefront restoration algorithm.

[0012] Step 5: Extract the interference fringe image from the second photodetector, perform Fourier transform on the interference fringe image to obtain the spectral image, use frequency domain filtering to obtain the +1st order term containing the reconstructed wavefront amplitude information, and then obtain the focusing plane light field distribution according to the scalar diffraction theory.

[0013] Step 6: Input the reconstructed wavefront from the Shaker-Hartmann wavefront measurement system and the reconstructed wavefront from the off-axis digital holographic wavefront measurement system into the MSWM-Net model to further obtain a high-confidence reconstructed wavefront.

[0014] The principle of this invention lies in a cross-scale composite wavefront sensing method based on a deep learning model. This method involves simultaneously acquiring distorted wavefronts using a Shak-Hartmann wavefront measurement system and an off-axis digital holographic wavefront measurement system. The acquired wavefronts are then fused using an MSWM-Net model to obtain a high-confidence reconstructed wavefront. The device includes a laser, fiber optic beam splitter, collimating lens, fast reflector, beam splitter, beam shrinking lens, microlens array, magnification matching system, photodetector, synchronization triggering device, and air outlet. These optical components constitute a composite wavefront sensing system, comprising both the Shak-Hartmann wavefront measurement system and the off-axis digital holographic wavefront measurement system.

[0015] Compared with the prior art, the present invention has the following advantages:

[0016] (1) This invention proposes a novel cross-scale composite wavefront sensing method that combines the system performance advantages of Shaker-Hartmann wavefront sensing and off-axis digital holographic wavefront sensing. When measuring aberrations involving different scales, this method overcomes the limitations of low phase accuracy in SHWFS wavefront sensing and the problem of the wavefront being wrapped in OADH measurements. The limitations of the cycle.

[0017] (2) This invention proposes an MSWM-Net model based on deep neural networks. This model is a high-efficiency multi-input single-output network structure composed of neural networks and convolutional attention mechanisms. By fusing the wavefronts measured by composite sensors, a high-confidence wavefront can be obtained. Compared with existing neural network models, it is easier to implement data fusion.

[0018] (3) When measuring wavefront aberrations containing different spatial frequencies, the present invention has higher wavefront restoration accuracy compared with existing methods. Attached Figure Description

[0019] Figure 1 This is a framework diagram of a cross-scale composite wavefront sensing method based on a deep learning model according to the present invention.

[0020] Figure 2 This diagram shows the experimental setup for a composite wavefront sensing system. In the diagram, 1 is a laser, 2 is an optical fiber beam splitter, 3-1 is a first collimating lens, 3-2 is a second collimating lens, 4 is a fast reflector, 5-1 is a first beam splitter, 5-2 is a second beam splitter, 6-1 is a first beam shrinking lens, 6-2 is a second beam shrinking lens, 7 is a microlens array, 8-1 is a first magnification matching system, 8-2 is a second magnification matching system, 9-1 is a first photodetector, 9-2 is a second photodetector, 10 is a synchronous triggering device, and 11 is an air outlet device.

[0021] Figure 3 A synchronous triggering system for lasers and photodetectors;

[0022] Figure 4 This is a schematic diagram of the MSWM-Net model, where Figure 4 The Arabic numerals listed are the number of channels, representing the depth or dimension of the feature map;

[0023] Figure 5 This is a reconstructed wavefront image obtained using a cross-scale composite wavefront sensing method based on a deep learning model. Detailed Implementation

[0024] To make the working principle and implementation process of the device of the present invention clearer, the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0025] Figure 1 This is a framework diagram of a cross-scale composite wavefront sensing method based on a deep learning model, as proposed in this invention. The specific implementation of this method first utilizes… Figure 2 A composite wavefront sensing system experimental device is used for wavefront measurement. The composite wavefront sensing system experimental device includes: a laser 1 and an optical fiber beam splitter 2 connected in sequence. After the optical fiber beam splitter 2, two branches are connected in parallel. The first branch consists of a first collimating lens tube 3-1, an air outlet device 11 for simulating atmospheric turbulence, a fast reflector 4, a first beam splitter 5-1, a first beam shrinking lens tube 6-1, a microlens array 7, a first magnification matching system 8-1, and a first photodetector 9-1. The second branch consists of a second collimating lens tube 3-2, a second beam splitter 5-2, a second beam shrinking lens tube 6-2, a second magnification matching system 8-2, and a second photodetector 9-2. The first photodetector 9-1 and the second photodetector 9-2 are connected to a synchronous triggering device 10, which is used to realize the synchronous triggering of the laser 1 and the first photodetector 9-1 and the second photodetector 9-2. The experimental setup utilizes a Shak-Hartmann wavefront measurement system and an off-axis digital holographic wavefront measurement system. A synchronous triggering device 10 synchronizes the photoelectric sensors corresponding to both systems, enabling them to acquire data at the same frequency. The acquired light spot array images and interference fringes are then further processed to obtain the light field distribution maps of the Shak-Hartmann reconstructed wavefront and the off-axis digital holography, respectively. Finally, these light field distribution maps are input into the MSWM-Net model.

[0026] In this embodiment of the invention, the laser wavelength is 1024 nm and the number of microlens arrays is 16. 16 (Ф4mm), sub-aperture size 0.24mm×0.24mm, microlens focal length 8.5mm, magnification matching system matching ratio The full aperture size is 3.8mm, and the pixel count is 15. The sampling frequency is 50 25000Hz. The photodetector used is the MEMERECAM ACS-1M60 from WeCare Corporation, as an example.

[0027] First, without turning on the air outlet in the optical path, the laser 1 is turned on to calibrate the system. After calibration, the air outlet 11 is turned on, and the composite sensing system is used to reconstruct the aberrant wavefront.

[0028] Before measuring the distorted wave, the synchronous triggering device 10 sends a synchronous triggering signal to the laser 1, the first photodetector 9-1, and the second photodetector 9-2 at a preset frequency. After receiving the triggering signal, the laser 1 emits a laser, and the first photodetector 9-1 and the second photodetector 9-2 begin to acquire images.

[0029] In one embodiment, the first magnification matching system 8-1 and the second magnification matching system 8-2 can be either a fixed imaging magnification system or a variable imaging magnification system.

[0030] In one embodiment, the air outlet device 11 is a device for simulating atmospheric turbulence, which can control the flow rate, or it may be a turbulence generator, etc.

[0031] The cross-scale composite wavefront sensing method based on deep learning models includes the following steps:

[0032] Step 1: First, the system is calibrated. This process includes: turning on the laser 1, the emitted beam passes through the fiber beam splitter 2 to obtain two identical emitted beams, namely the first beam O and the second beam R. The first beam O passes through the first collimating lens tube 3-1, the fast reflector 4, and the first beam splitter 5-1, and then enters the first beam shrinking lens tube 6-1, before exiting to the microlens array 7. After being split by the microlens array 7, the first beam O passes through the first magnification matching system 8-1 and is imaged in the form of a calibration spot array on the first photodetector 9-1, forming a calibration spot image. The second beam R passes through the second collimating lens tube 3-2 and then through the second beam splitter 5-2, interfering with the first beam O that passed through the first beam splitter 5-1. After passing through the second beam shrinking lens tube 6-2, it is imaged on the second photodetector 9-2, forming an interference fringe image.

[0033] When measuring aberrations, the synchronous triggering device 10 activates at a preset frequency, thereby synchronously triggering signals to the laser 1, the first photodetector 9-1, and the second photodetector 9-2. For example... Figure 3As shown, after receiving the trigger signal, laser 1 emits a laser pulse. Upon detecting the falling edge, the first photodetector 9-1 and the second photodetector 9-2 initiate the electronic shutter pulse, initiating the exposure time. Upon detecting the rising edge, the first photodetector 9-1 and the second photodetector 9-2 begin charge transfer, ending the exposure and simultaneously outputting an image signal to the acquisition card.

[0034] Step 2: Extract the calibration spot image from the first photodetector 9-1 and the interference fringe image from the second photodetector 9-2;

[0035] Step 3: Turn on the air outlet device 11. When laser 1 is turned on, the emitted beam passes through fiber optic beam splitter 2 to obtain two identical emitted beams, namely the first beam O and the second beam R. The first beam O is emitted through the first collimating lens tube 3-1 to the air outlet device 11. At this time, the first beam O has aberrations. Then it passes through the fast reflector 4 and the first beam splitter 5-1. The first beam splitter 5-1 splits the first beam O into a third beam O_1 and a fourth beam O_2. The fourth beam O_2 passes through the second beam splitter 5-2. The third beam O_1 enters the first beam shrinking lens tube 6-1 and then exits to the microlens array 7. After being split by the microlens array 7, the third beam O_1 is imaged in the first photodetector 9-1 in a dot matrix form with spatial offset relative to the calibration position by the first magnification matching system 8-1. The second beam R passes through the second collimating lens tube 3-2 and then through the second beam splitter 5-2, interfering with the fourth beam O_2. After passing through the second beam shrinking lens tube 6-2, it is imaged in the second photodetector 9-2 to form an interference fringe image with aberrations.

[0036] Step 4: Extract the distorted spot array image on the first photodetector 9-1, calculate the offset of the distorted spot relative to the calibration position using spot positioning technology, calculate the wavefront slope using the offset, and then reconstruct the distorted wavefront using the wavefront restoration algorithm.

[0037] In one embodiment, the spot localization technique includes the weighted centroid method, the threshold centroid method, the matched filtering method, the registration algorithm, or any other method that can locate the spot position.

[0038] The following describes the process of calculating the wavefront slope:

[0039] Microlens array in Complex amplitude of the light field on the focal plane of the sub-aperture position for:

[0040] ,

[0041] in, , This corresponds to the position information of the sub-aperture. The complex amplitude corresponding to the first beam O in step one. , This is the sub-aperture calibration position information. It is the imaginary unit. The wavelength of the laser. for Figure 1 Focal length and wavenumber of microlens array 7 .

[0042] The centroid position of the focused spot in the x-direction is:

[0043] ,

[0044] in, To focus the centroid position of the light spot in the x-direction, for conjugate, To take the real part of the complex number.

[0045] The centroid position of the focused spot in the y-direction is:

[0046] ,

[0047] in, To focus the centroid of the light spot along the y-direction, for conjugate, To take the real part of the complex number.

[0048] Wavefront slopes in the x and y directions of the incident wavefront:

[0049] ,

[0050] in, , The slopes of the incident wavefront in the x and y directions are given by the wavefront slope. , Indicates the first Location of the centroid of the focused light spot from the incident light wave of the aperture. and This indicates the position of the centroid of the focused light spot.

[0051] The distorted wavefront is then reconstructed using wavefront reconstruction algorithms. In one embodiment, wavefront reconstruction algorithms include mode-based methods, region-based methods, and other methods that can be used to reconstruct the wavefront from sub-spot array data of a Shaker-Hartmann wavefront sensor. The Shaker-Hartmann wavefront sensor uses the mode-based method for wavefront reconstruction, further obtaining the reconstructed wavefront from the Shaker-Hartmann wavefront measurement system. The reconstructed Zernike polynomial coefficients are:

[0052] ,

[0053] Where D is the number of effective sub-apertures; R is the wavefront restoration matrix, which is the restoration matrix for calculating aberration mode coefficients based on the centroid shift or slope data of the spot within the sub-aperture. The matrix can be generated in advance as a system configuration; M is the order of the reconstructed Zernike polynomial. for( , ) corresponds to the wavefront slope matrix; a is the reconstructed Zernike polynomial coefficient.

[0054] Step 5: Extract the interference fringe image on the second photodetector 9-2, perform Fourier transform on the interference fringe image to obtain the spectral image, use frequency domain filtering to obtain the +1 order term containing the reconstructed wavefront amplitude information, and then obtain the light field distribution of the focusing plane according to the scalar diffraction theory, and further obtain the reconstructed wavefront of the Shakhartmann wavefront measurement system.

[0055] Step Six: Input the reconstructed wavefront from the Shackleton-Hartmann wavefront measurement system and the reconstructed wavefront from the off-axis digital holographic wavefront measurement system into, for example... Figure 4 In the MSWM-Net model shown, high-confidence reconstructed wavefronts are further obtained, such as... Figure 5 As shown.

[0056] The MSWM-Net model according to an embodiment of the present invention is described below. The MSWM-Net model is a high-efficiency multi-input single-output network structure composed of a neural network and a convolutional attention mechanism. The framework of this model is based on the typical U-Net structure, including convolutional layers, max-pooling layers, upsampling layers (Convtranspose2d), and a convolutional attention mechanism (CBAM) module. Each convolutional layer is followed by a batch normalization layer and a LeakyReLu activation function to accelerate training. Convtranspose2d halves the number of feature channels and concatenates them with corresponding cropped features from the downsampling blocks. CBAM generates image features with attention mechanism weights from both channel and spatial dimensions to improve the network's feature extraction capability. Figure 4 The general framework of the model is shown, where the listed Arabic numerals represent the number of channels, indicating the depth of the feature maps. Downsampling, accompanied by an increase in the number of channels, better represents the attributes and features of the input image, thus improving segmentation accuracy. Upsampling, by reducing the number of channels, allows the network to reduce computational complexity while maintaining segmentation performance, thereby improving inference speed.

[0057] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any transformations or substitutions that can be conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of the present invention.

Claims

1. A cross-scale composite wavefront sensing method based on a deep learning model, characterized in that, Wavefront measurement is performed using a composite wavefront sensing system experimental device, which includes: a laser (1) and an optical fiber beam splitter (2) connected in sequence. Two branches are connected in parallel after the optical fiber beam splitter (2), consisting of a first collimating lens tube (3-1), an air outlet device (11) for simulating atmospheric turbulence, a fast reflector (4), a first beam splitter (5-1), a first beam-shrinking lens tube (6-1), a microlens array (7), a first magnification matching system (8-1), and a first photodetector (…). The first branch consists of a second collimating lens tube (3-2), a second beam splitter (5-2), a second beam shrinking lens tube (6-2), a second magnification matching system (8-2), and a second photodetector (9-2). The first photodetector (9-1) and the second photodetector (9-2) are connected to a synchronous triggering device (10). The synchronous triggering device (10) is used to realize the synchronous triggering of the laser (1) and the first photodetector (9-1) and the second photodetector (9-2). The cross-scale composite wavefront sensing method based on deep learning models includes the following steps: Step 1: First, the system is calibrated. This process includes: turning on the laser (1), the emitted beam passes through the fiber beam splitter (2) to obtain two identical emitted beams, namely the first beam and the second beam. The first beam passes through the first collimating lens tube (3-1), the fast reflector (4) and the first beam splitter (5-1) and enters the first beam shrinking lens tube (6-1), and then exits to the microlens array (7). After being split by the microlens array (7), the first beam passes through the first magnification matching system (8-1) and is imaged in the form of a calibration spot array on the first photodetector (9-1) to form a calibration spot image. The second beam passes through the second collimating lens tube (3-2) and then through the second beam splitter (5-2), interfering with the first beam that passed through the first beam splitter (5-1). After passing through the second beam shrinking lens tube (6-2), it is imaged on the second photodetector (9-2) to form an interference fringe image. Step 2: Extract the calibration spot image from the first photodetector (9-1) and the interference fringe image from the second photodetector (9-2); Step 3: Turn on the air outlet device (11) and the laser (1). The emitted beam passes through the fiber optic beam splitter (2) to obtain two identical emitted beams, namely the first beam and the second beam. The first beam is emitted through the first collimating lens tube (3-1) and then to the air outlet device (11). At this time, the first beam has aberrations. Then, it passes through the fast reflector (4) and the first beam splitter (5-1). The first beam splitter (5-1) splits the first beam into a third beam and a fourth beam. The fourth beam passes through the second beam splitter (5-2), and the third beam enters the second beam. The beam enters the first beam-shrinking tube (6-1) and exits into the microlens array (7). After being split by the microlens array (7), the third beam is imaged in the first photodetector (9-1) in a dot matrix form with spatial offset relative to the calibration position through the first magnification matching system (8-1). The second beam passes through the second collimating tube (3-2) and then through the second beam splitter (5-2), where it interferes with the fourth beam. After passing through the second beam-shrinking tube (6-2), it is imaged in the second photodetector (9-2) to form an interference fringe image with aberrations. Step 4: Extract the distorted spot image on the first photodetector (9-1), calculate the offset of the distorted spot relative to the calibration position using spot positioning technology, calculate the wavefront slope using the offset, and then reconstruct the distorted wavefront using the wavefront restoration algorithm. Step 5: Extract the interference fringe image on the second photodetector (9-2), perform Fourier transform on the interference fringe image to obtain the spectral image, use frequency domain filtering to obtain the +1 order term containing the reconstructed wavefront amplitude information, and then obtain the focusing plane light field distribution according to the scalar diffraction theory; Step 6: Input the reconstructed wavefront from the Shaker-Hartmann wavefront measurement system and the reconstructed wavefront from the off-axis digital holographic wavefront measurement system into the MSWM-Net model to further obtain a high-confidence reconstructed wavefront.

2. The cross-scale composite wavefront sensing method based on a deep learning model according to claim 1, characterized in that, The first magnification matching system (8-1) and the second magnification matching system (8-2) are either fixed magnification systems or variable magnification systems.

3. The cross-scale composite wavefront sensing method based on a deep learning model according to claim 1, characterized in that, The air outlet device (11) is a device for simulating atmospheric turbulence, or a turbulence generator.

4. The cross-scale composite wavefront sensing method based on a deep learning model according to claim 1, characterized in that, Spot localization techniques include weighted centroid method, threshold centroid method, matched filtering method, and registration algorithm.

5. The cross-scale composite wavefront sensing method based on a deep learning model according to claim 1, characterized in that, Wavefront reconstruction algorithms include either the pattern method or the region method.

6. The cross-scale composite wavefront sensing method based on a deep learning model according to claim 1, characterized in that, The MSWM-Net model is a high-efficiency multi-input single-output network structure composed of neural networks and convolutional attention mechanisms.

Images