Laser space-variant PSF aberration elimination system and method based on artificial intelligence

Abstract

The invention relates to the technical field of aberration elimination, and discloses a laser space-variant PSF aberration elimination system and method based on artificial intelligence, and the system comprises a first module which is used for calculating an envelope energy curve; the second module is used for constructing a center hollowness index heat map; the third module is used for calculating a backfill intention weight field; the fourth module is used for calculating a space-variant kernel feasible set; the fifth module is used for establishing a spatial variation convolution observation model; the sixth module is used for updating the clear image and the space-variant kernel; the seventh module recalculates a center hollowness index heat map; and the eighth module is used for outputting the restored image and the space-variant kernel. According to the invention, the problem of core energy collapse caused by laser space-variant PSF aberration is eliminated through the center hollowness index; a space-variant kernel feasible set containing non-negative, unit sum, envelope energy monotonous non-decreasing and core energy base line constraint is constructed, and a backfill intention weight field is combined, so that artifacts caused by undifferentiated correction are avoided.

Description

Technical Field

[0001] This invention relates to the field of aberration correction technology, and more specifically, to an artificial intelligence-based laser spatially variable PSF aberration correction system and method. Background Technology

[0002] Laser imaging technology, with its high resolution and long-range detection capabilities, is widely used in high-precision target recognition, laser communication, remote sensing and other fields. Its imaging quality directly determines the accuracy of subsequent signal processing and target interpretation. However, in practical applications, lasers are susceptible to interference from the external environment during transmission. At the same time, the optical system itself has spatially dependent aberrations, which cause the point spread function to change with spatial position, forming spatially variable PSF aberrations.

[0003] The generation of this spatial variation characteristic mainly stems from two aspects: First, the turbulence causes the intensity of the receiving surface to fluctuate randomly, and the small window or pixel equivalent small aperture makes it easier to observe deep fading, resulting in the energy of the core area being hollowed out, presenting a hollow center phenomenon; Second, the aperture, focal length and other parameters of the optical system have spatial dependence, and the diffused spot shape and energy distribution formed by point light sources at different field positions after passing through the system are different.

[0004] Existing aberration elimination methods have significant limitations: some methods use general priors for indiscriminate correction, which cannot specifically address the core energy collapse problem; some rely on phase information or neural network models, lacking physical interpretability and easily introducing artifacts; and some methods do not fully consider energy conservation and core energy baseline constraints, resulting in restoration results that do not conform to the physical laws of optical imaging. These problems make it difficult for existing technologies to accurately eliminate spatially varied PSF aberrations while ensuring physical rationality, and cannot meet the performance requirements of high-precision laser imaging scenarios. Therefore, there is an urgent need for a directional, interpretable, and physically constrained aberration elimination scheme. Summary of the Invention

[0005] This invention provides an artificial intelligence-based laser spatially variable PSF aberration elimination system and method, which solves the technical problems mentioned in the background.

[0006] This invention provides an artificial intelligence-based laser spatially variable PSF aberration elimination system, comprising: The first module is used to perform sliding window processing on the normalized image sequence, calculate the intensity centroid for each sliding window, and obtain the envelope energy curve with the intensity centroid as the center. The second module is used to determine the minimum radius through a small energy threshold, combine it with the diffraction-limited radius to determine the central void index, and construct a central void index heat map accordingly. The third module is used to establish a radial window smaller than the diffraction limit radius and to map the central void index into a backfill intention weight field. The fourth module is used to define a feasible set of space-variable kernels for each sliding window, and to impose constraints that are non-negative, unity sum, monotonically non-decreasing envelope energy, and not lower than a small energy threshold at the diffraction-limited radius. The fifth module is used to establish a spatial variation convolutional observation model using a partitioned consistent window. It constructs an objective function that includes a data consistency term, a central envelope energy backfill term, and a total variation regularization term. The backfill intention weight field is used to weight the function and penalize insufficient envelope energy only within the range of the diffraction limit radius. The sixth module is used to iteratively update the clear image and the spatially variable kernel by alternating minimization, and to project to the feasible set of spatially variable kernels after each update; The seventh module is used to recalculate the heat map of the central hollowness index based on the restored image, determine the convergence amount accordingly, and control the iteration to stop or increase the backfill intention weight field at the corresponding position based on the convergence amount. The eighth module is used to output the restored image and the spatial variation kernel, and to perform energy conservation checks and envelope energy consistency checks.

[0007] Furthermore, energy normalization is performed on each frame of the image within the region of interest, so that the sum of the intensity values ​​of all pixels within the region of interest equals 1, thus obtaining a normalized image. A sliding window is set up, which is configured with a binary mask, and the pixel values ​​of the binary mask are only 0 or 1. The sum of the intensity values ​​of all pixels covered by the binary mask of the sliding window is calculated to obtain the total intensity of the window. The intensity value of each pixel covered by the binary mask of the sliding window is divided by the total intensity of the window to obtain the normalized intensity of the window for each corresponding pixel.

[0008] Furthermore, using the normalized intensity of the small window for each pixel as a weight, the coordinates of all pixels covered by the binary mask of the sliding small window are weighted and summed. The weighted sum is then divided by the sum of the normalized intensities of the small windows for all pixels to obtain the intensity centroid. Using the intensity centroid as the center, the sum of the normalized intensities of the small windows for all pixels within each radius range is calculated to form an envelope energy curve. A small energy threshold with a value greater than 0 and less than 1 is set, and the minimum radius that makes the value of the envelope energy curve reach the small energy threshold is found. A diffraction limit radius is set, and the found minimum radius is divided by the diffraction limit radius to obtain the center hollowness index. The median of the center hollowness indices obtained for the same sliding small window in all frame images is taken, and the medians corresponding to all sliding small windows together constitute the center hollowness index heatmap.

[0009] Furthermore, the numerical value corresponding to each sliding window in the central hollowness index heatmap is extracted, and 1 is subtracted from this value to obtain the difference for each sliding window. The difference for each sliding window is multiplied by the scheduling intensity coefficient to obtain the first product for each sliding window. The first product for each sliding window is multiplied by the radial window value corresponding to different radial distance variables to obtain the second product for each sliding window at different radial distances. The second product for each sliding window at different radial distances is added to 1 to obtain the backfilling intention weight for each sliding window at different radial distances. The backfilling intention weights of all sliding windows at different radial distances together constitute the backfilling intention weight field. radial window The calculation formula is as follows: Where r represents the radial distance variable, denoted by the diffraction-limiting radius, and exp denotes an exponential function with the natural constant as its base.

[0010] Furthermore, firstly, each value of the kernel to be projected is compared with 0, and the larger value is taken as the value of the intermediate kernel. The sum of all the values ​​of the intermediate kernel is calculated, and each value of the intermediate kernel is divided by the sum to obtain the normalized intermediate kernel. Define a discrete radius sequence, calculate the sum of the values ​​of the normalized intermediate kernel within each discrete radius range to obtain the original cumulative energy sequence; for each discrete radius interval, calculate the scaling factor of that interval, which is the difference between the corresponding intervals in the new cumulative energy sequence divided by the sum of the values ​​of the normalized intermediate kernels in that interval. Multiply the value of the normalized intermediate kernel in each discrete radius interval by the scaling factor of the corresponding interval to obtain the monotonic kernel; calculate the envelope energy value of the monotonic kernel at the diffraction-limited radius. If the value is less than the small energy threshold, calculate the gap value, which is the small energy threshold minus the envelope energy value. Define a core region and an outer region. The core region is the area with the center of the sliding window as the center and the diffraction limit radius as the radius. The outer region is the area outside the core region. Calculate the sum of the number of pixels in the core region and the number of pixels in the outer region. For the monotonic kernel values ​​within the core region, each value is increased by the gap value divided by the number of pixels in the core region; for the monotonic kernel values ​​within the outer region, each value is multiplied by 1 and subtracted from the gap value divided by the sum of the values ​​in the outer region to obtain the spatially variable kernel.

[0011] Furthermore, each partition's consistent window is multiplied point-by-point by the convolution result of the corresponding spatial variation kernel and the clear image, and then the point-by-point multiplication results of all partitions are summed to obtain the observation model corresponding to the normalized image. The partition consistent window satisfies that the sum of the values ​​of all partition consistent windows is equal to 1 and each value is greater than or equal to 0. An objective function is constructed, which consists of three terms: the first term is the squared L2 norm of the difference between the normalized image and the spatial variation convolution result; the second term is the product of the squared negative part of the difference between the backfill intention weight field and the reference envelope energy and the spatial variation kernel envelope energy within the diffraction limit radius for each partition, and then the integral result of all partitions is multiplied by the first weight coefficient; the third term is the total variation of the clear image multiplied by the second weight coefficient.

[0012] Furthermore, the sharp image is iteratively updated by alternating minimization, including the following steps: First, the convolution result of the consistent window of each partition and the spatial variation kernel after flipping is calculated. This convolution result is multiplied point by point by the result of the normalized image divided by the spatial variation convolution result plus the numerical stability constant. Then, the point-by-point multiplication results of all partitions are added together. Next, the sum of the convolution result of the consistent window of each partition and the spatial variation kernel after flipping plus the numerical stability constant is calculated. The former is divided by the latter and then multiplied point by point by the current sharp image to obtain the updated intermediate sharp image. Then, the sharp image that minimizes the sum of the squared L2 variation regularization term of the difference between the intermediate sharp image and the sharp image to be updated is selected as the new sharp image. The spatially variable kernel is iteratively updated by alternating minimization, including the following steps: First, the residual is calculated, which is the difference between the normalized image and the spatially variable convolution result; then, for each partition, the pointwise product of the partition consistency window and the residual is calculated, and the convolution result after flipping the clear image is multiplied by negative two, plus the product of the weight coefficient and the gradient of the envelope energy penalty term of the spatially variable kernel. This result is multiplied by the step size and subtracted from the current spatially variable kernel to obtain an intermediate spatially variable kernel; finally, the intermediate spatially variable kernel is projected onto the feasible set of spatially variable kernels. This projection sequentially implements non-negativity constraints, unit sum constraints, envelope energy monotonicity constraints based on monotonic regression, and small energy threshold constraints at the diffraction limit radius to obtain a new spatially variable kernel.

[0013] Furthermore, the iteration is controlled based on the convergence value, including the following steps: Step S201: Using the restored image as the calculation object, calculate the envelope energy curve according to the process of central hollowness index measurement and spatial variation mapping, determine the minimum radius that reaches the small energy threshold, divide the minimum radius by the diffraction limit radius to obtain the instantaneous central hollowness index, take the median value of the instantaneous central hollowness index in the time dimension, and obtain the restored central hollowness index heat map. Step S202: Calculate the convergence factor, which is the median of the absolute values ​​of the differences between the hollowness index of the restoration center of all sliding windows and 1; calculate the data consistency residual, which is the squared L2 of the difference between the normalized image and the spatial variation convolution result. Step S203: Set the monitoring interval. At each iteration interval, calculate the absolute value of the difference between the current convergence and the convergence at the previous monitoring time, as well as the absolute value of the difference between the current data consistency residual and the data consistency residual at the previous monitoring time. Step S204: Determine whether to stop the iteration. When the current convergence is less than or equal to the convergence at the previous monitoring time, the absolute value of the difference between the convergences is less than or equal to the numerical tolerance, and the current data consistency residual is less than or equal to the data consistency residual at the previous monitoring time, the absolute value of the difference between the data consistency residuals is less than or equal to the numerical tolerance, the iteration is stopped. Step S205: If there are sliding windows with a hollowness index of the restoration center greater than 1, update the backfilling intention weight field corresponding to these sliding windows. The update method is to multiply the original backfilling intention weight field by 1 and add the product of the gain coefficient and the hollowness index of the restoration center minus 1. The value of the gain coefficient is greater than 0 and less than or equal to 1. Step S206: Repeat steps S201 to S205 until the stopping iteration condition is met.

[0014] Furthermore, an energy conservation check is performed. The sum of the intensity values ​​corresponding to all pixel coordinates and all frame numbers in the input image is calculated to obtain the total input intensity. The sum of the intensity values ​​corresponding to all pixel coordinates and all frame numbers in the restored image is calculated to obtain the total output intensity. The absolute value of the difference between the total input intensity and the total output intensity is calculated as the energy difference. To perform an envelope energy consistency check, for each sliding window's spatially variable kernel, with the center of the sliding window as the center, calculate the sum of the spatially variable kernel values ​​corresponding to all pixel coordinates within each radius range to form an envelope energy curve; at the diffraction-limited radius, check whether the value of the envelope energy curve is not less than the low energy threshold. To verify the consistency of the restored center hollowness index, the restored image was used as the calculation object. Following the process of center hollowness index measurement and spatial variation mapping, the envelope energy curve was calculated to determine the minimum radius that reaches the small energy threshold. The minimum radius was divided by the diffraction limit radius to obtain the instantaneous center hollowness index. The median of the instantaneous center hollowness index in the time dimension was taken to obtain the heat map of the restored center hollowness index. The median of the absolute values ​​of the differences between the restored center hollowness index of all sliding windows and one was calculated.

[0015] This invention provides an artificial intelligence-based method for eliminating laser spatially variable PSF aberrations, comprising the following steps: Step S301: Perform sliding window processing on the normalized image sequence, calculate the intensity centroid for each sliding window, and obtain the envelope energy curve with the intensity centroid as the center. Step S302: Determine the minimum radius through the small energy threshold, determine the central voidness index by combining the diffraction-limited radius, and construct a central voidness index heat map accordingly. Step S303: Establish a radial window smaller than the diffraction limit radius and map the central hollowness index to a backfill intention weight field; Step S304: Define a feasible set of empty variable kernels for each sliding window, and apply constraints that are non-negative, unity sum, monotonically non-decreasing envelope energy, and not lower than the small energy threshold at the diffraction limit radius; Step S305: A spatial variation convolutional observation model is established using a partitioned consistent window. An objective function is constructed that includes a data consistency term, a central envelope energy backfill term, and a total variation regularization term. The backfill intention weight field is used to weight the function and penalize insufficient envelope energy only within the range of less than the diffraction limit radius. Step S306: Iteratively update the clear image and the spatially variable kernel by alternating minimization, and project the image to the feasible set of the spatially variable kernel after each update. Step S307: Recalculate the heat map of the central hollowness index based on the restored image, determine the convergence amount accordingly, and control the iteration to stop or increase the backfill intention weight field at the corresponding position based on the convergence amount. Step S308: Output the restored image and the spatial variation kernel, and perform energy conservation check and envelope energy consistency check.

[0016] The beneficial effects of this invention are as follows: This invention eliminates the core energy collapse problem caused by laser void-variable PSF aberration through the central hollowness index; by constructing a feasible set of void-variable kernels containing non-negative, unit sum, monotonically non-decreasing envelope energy and core energy baseline constraints, combined with a backfill intention weight field to directionally guide the energy replenishment of the core region, artifacts caused by indiscriminate correction are avoided; when alternately minimizing and updating clear images and void-variable kernels, they are synchronously projected onto the feasible set, and with closed-loop control iteration, the restoration results are ensured to fit the observation data and the core hollowness phenomenon is effectively eliminated; at the same time, the energy conservation and envelope energy consistency verification links further ensure that the output results are reliable and reproducible, and the overall results meet the requirements of high-precision laser imaging scenarios for the directionality, interpretability and physical completeness of aberration elimination. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a laser spatially variable PSF aberration elimination system based on artificial intelligence according to the present invention; Figure 2 This is a flowchart of an artificial intelligence-based laser spatially variable PSF aberration elimination method according to the present invention.

[0018] In the diagram: Module 101, Module 202, Module 303, Module 404, Module 505, Module 606, Module 707, Module 808. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0020] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" indicate that the element or object preceding the term encompasses the elements or objects listed following the term and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0021] like Figures 1-2 As shown, an artificial intelligence-based laser spatially variable PSF aberration elimination system includes: The first module 101 is used to perform sliding window processing on the normalized image sequence, calculate the intensity centroid for each sliding window, and obtain the envelope energy curve with the intensity centroid as the center. The second module 102 is used to determine the minimum radius through a small energy threshold, combine it with the diffraction-limited radius to determine the central void index, and construct a central void index heat map accordingly. The third module 103 is used to establish a radial window smaller than the diffraction limit radius and to map the central void index to a backfill intention weight field. The fourth module 104 is used to define a feasible set of space-variable kernels for each sliding window, and to impose constraints that are non-negative, unity sum, monotonically non-decreasing envelope energy, and not lower than a small energy threshold at the diffraction-limited radius. The fifth module 105 is used to establish a spatial variation convolutional observation model using a partitioned consistent window, construct an objective function that includes a data consistency term, a central envelope energy backfill term, and a total variation regularization term, and weights it with a backfill intention weight field and only penalizes insufficient envelope energy within the range of less than the diffraction limit radius. The sixth module 106 is used to iteratively update the clear image and the spatially variable kernel by alternating minimization, and to project to the feasible set of the spatially variable kernel after each update of the spatially variable kernel; The seventh module 107 is used to recalculate the heat map of the central hollowness index based on the restored image, determine the convergence amount accordingly, and control the iteration to stop or increase the backfill intention weight field at the corresponding position based on the convergence amount. The eighth module 108 is used to output the restored image and the spatial variation kernel, and to perform energy conservation check and envelope energy consistency check.

[0022] It should be noted that the spatially varied point spread function (PSF) of a laser is a characteristic of the point spread function in a laser optical system that varies with spatial position. Its morphological parameters change with the field of view position, and the diffused light spots formed by point sources at different spatial positions after passing through the system differ. This variation stems from the spatially dependent aberrations of the optical system or the influence of the external environment. Laser spatially varied PSF aberration elimination aims to optimize the imaging and transmission performance of laser optical systems. Through relevant technical means, it ensures that the diffused light spots formed by point sources at different spatial positions after passing through the system are consistent, reducing the impact of spatial position on imaging sharpness, ensuring the integrity of the laser signal during transmission or imaging, providing a stable optical foundation for subsequent signal processing or target recognition, and meeting the performance requirements of laser systems in high-precision application scenarios.

[0023] In one embodiment of the present invention, energy normalization processing is performed on each frame of image within the region of interest, so that the sum of the intensity values ​​of all pixels in the region of interest equals 1, thereby obtaining a normalized image; a sliding window is set, which is configured with a binary mask, the pixel values ​​of which are only 0 or 1; the sum of the intensity values ​​of all pixels covered by the binary mask of the sliding window is calculated to obtain the total intensity of the window; the intensity value of each pixel covered by the binary mask of the sliding window is divided by the total intensity of the window to obtain the normalized intensity of the window for each corresponding pixel.

[0024] It should be noted that the region of interest (ROI) needs to be determined around the imaging target or the effective imaging range. This can be done by manually selecting the region or by automatically identifying continuous high-signal areas based on the image's grayscale contrast. This ensures that the region completely covers the core imaging part that needs to be analyzed and avoids interference from irrelevant backgrounds. The intensity value of all pixels refers to the quantized result of light intensity detection for each pixel within the ROI. This is the direct capture and digital representation of the laser signal reflected or emitted by the target by the imaging system. The size of the sliding window is determined based on the imaging resolution and the spatial coherence scale of the turbulence. It is typically set to 8×8 pixels to 32×32 pixels. The window must be able to capture local spatial variation characteristics and contain enough pixels to support intensity statistics and centroid calculation. The size of the binary mask is exactly the same as that of the sliding window. The pixel positions covered by the window in the mask are set to 1, and the positions not covered by the window are set to 0. This is used to accurately define the spatial range of the window. Each pixel covered by the sliding window corresponds to a normalized intensity of the window. Pixels in the non-covered areas do not have this value.

[0025] In one embodiment of the present invention, the coordinates of all pixels covered by the binary mask of the sliding window are weighted and summed using the normalized intensity of the small window of each pixel as the weight. The weighted sum is then divided by the sum of the normalized intensities of the small windows of all pixels to obtain the intensity centroid. Using the intensity centroid as the center, the sum of the normalized intensities of the small windows of all pixels within each radius range is calculated to form an envelope energy curve. A small energy threshold with a value greater than 0 and less than 1 is set, and the minimum radius that makes the value of the envelope energy curve reach the small energy threshold is found. A diffraction limit radius is set, and the found minimum radius is divided by the diffraction limit radius to obtain the central hollowness index. The median of the central hollowness indices obtained for the same sliding window in all frame images is taken, and the medians corresponding to all sliding windows together constitute a central hollowness index heatmap.

[0026] It should be noted that the intensity centroid is the weighted center of the intensity distribution within the sliding window, and its position accurately represents the core landing point of the intensity distribution within the window. Different radius ranges are set with the intensity centroid as the starting point and increase at equal intervals. The interval size matches the imaging pixel scale, typically ranging from 0.5 pixels to 1 pixel. The maximum radius must cover the main range of the intensity distribution within the window to ensure that the envelope energy curve can accumulate completely from 0 to 1. The envelope energy curve is a curve that reflects the cumulative change of intensity energy within the window with radial distance. The horizontal axis represents the radial distance with the intensity centroid as the center, i.e., the radius value, which intuitively shows the range extending outward from the intensity center. The vertical axis represents the sum of the normalized intensities of all pixels within the corresponding radius, with a value range between 0 and 1, directly reflecting the proportion of energy contained within that radius.

[0027] It should be noted that the minimum radius is the minimum radial distance at which the envelope energy curve reaches a preset low energy threshold. It is a key parameter for quantifying the degree of energy concentration in the central region, and preferably, the preset low energy threshold is 5%. The diffraction-limited radius is the radius of the core region of the light spot after imaging from a point source when there is no aberration interference. It is determined by the system aperture and the working wavelength and is a basic measure of the system's inherent resolution. Its setting needs to be calculated based on the optical system parameters. The specific calculation method is 1.22 multiplied by the working wavelength and then divided by the system aperture. This value is a standard reference value in optical design and imaging engineering. The central void index is the ratio of the minimum radius to the diffraction-limited radius. It is used to quantify the degree of energy collapse in the central region of the intensity distribution within the small window. The central void index heatmap is an image formed by arranging the central void indices corresponding to all sliding windows one by one according to their spatial positions. The value of each spatial position in the image corresponds to the central void index of the sliding window at that position. The numerical distribution intuitively presents the spatial differences in central energy collapse at different positions in the entire imaging area, providing a spatial distribution basis for subsequent targeted correction of spatial variation aberrations.

[0028] In one embodiment of the present invention, the value corresponding to each sliding window in the heat map of the central hollowness index is extracted, and 1 is subtracted from the value to obtain the difference corresponding to each sliding window. Multiply the difference corresponding to each sliding window by the scheduling intensity coefficient to obtain the first product corresponding to each sliding window; multiply the first product corresponding to each sliding window by the radial window value corresponding to different radial distance variables to obtain the second product corresponding to each sliding window at different radial distances. Add 1 to the second product corresponding to each sliding window at different radial distances to obtain the backfilling intention weight of each sliding window at different radial distances. The backfilling intention weights of all sliding windows at different radial distances together constitute the backfilling intention weight field. radial window The calculation formula is as follows: Where r represents the radial distance variable, denoted by the diffraction-limiting radius, and exp denotes an exponential function with the natural constant as its base.

[0029] It should be noted that the radial distance variable is the radial distance relative to the intensity centroid of the sliding window, and can be set to 0.2 times, 0.5 times, 0.8 times, and 1 times the diffraction limit radius, respectively, to cover the range from the core region to the diffraction limit radius, ensuring a refined weight distribution within the core region. The radial window is a function that monotonically decays with increasing radial distance. The scheduling intensity coefficient is used to control the intensity of the backfill intention as the central hollowness index changes, and its value ranges from 0.5 to 1. Preferably, the scheduling intensity coefficient is set to 0.7. The backfill intention weight field is the set of backfill intention weights for all sliding windows at different radial distances. Each weight value reflects the priority of the corresponding window in pulling energy back to the center at the corresponding radial position. The larger the value, the higher the priority. Subsequent optimization processes will use this weight field to make directional corrections to the energy distribution in the core region.

[0030] In one embodiment of the present invention, each value of the kernel to be projected is first compared with 0, and the larger value is taken as the value of the intermediate kernel. The sum of all the values ​​of the intermediate kernel is calculated, and each value of the intermediate kernel is divided by the sum to obtain the normalized intermediate kernel. Define a discrete radius sequence, calculate the sum of the values ​​of the normalized intermediate kernel within each discrete radius range to obtain the original cumulative energy sequence; for each discrete radius interval, calculate the scaling factor of that interval, which is the difference between the corresponding intervals in the new cumulative energy sequence divided by the sum of the values ​​of the normalized intermediate kernels in that interval. Multiply the value of the normalized intermediate kernel in each discrete radius interval by the scaling factor of the corresponding interval to obtain the monotonic kernel; calculate the envelope energy value of the monotonic kernel at the diffraction-limited radius. If the value is less than the small energy threshold, calculate the gap value, which is the small energy threshold minus the envelope energy value. Define a core region and an outer region. The core region is the area with the center of the sliding window as the center and the diffraction limit radius as the radius. The outer region is the area outside the core region. Calculate the sum of the number of pixels in the core region and the number of pixels in the outer region. For the monotonic kernel values ​​within the core region, each value is increased by the gap value divided by the number of pixels in the core region; for the monotonic kernel values ​​within the outer region, each value is multiplied by 1 and subtracted from the gap value divided by the sum of the values ​​in the outer region to obtain the spatially variable kernel.

[0031] It should be noted that the kernel to be projected is the estimated value of the point spread function (PSF). The discrete radius sequence is based on the pixel grid of the image and is set in an equally spaced incrementing manner, for example, starting from 0 and with an interval of 0.5 pixels or 1 pixel, until it covers the maximum radial range within the sliding window (which must include the diffraction limit radius). The discrete radius range represents the interval between two adjacent discrete radius values. The scaling factor is used to adjust the spatially variable kernel values ​​within the corresponding discrete radius interval so that the adjusted envelope energy sequence satisfies the monotonically non-decreasing constraint. The core region and the outer region are defined. Under the premise of ensuring the conservation of spatially variable kernel energy (summing to 1) and non-negativity, part of the energy in the outer region is reasonably allocated to the core region to meet the constraint that the envelope energy at the diffraction limit radius is not lower than the small energy threshold.

[0032] It should be noted that the spatially variable kernel represents the point spread function that varies with spatial position within the sliding window, used to describe the blurring characteristics of laser imaging in this local area. Each spatially variable kernel in the feasible set satisfies four conditions: all values ​​of the spatially variable kernel are greater than or equal to 0, the sum of all values ​​of the spatially variable kernel is equal to 1, the envelope energy curve is monotonically non-decreasing as the radius increases, and the envelope energy curve value at the diffraction-limited radius is not lower than the low energy threshold.

[0033] In one embodiment of the present invention, each partition consistency window is multiplied point-by-point by the convolution result of the corresponding spatial variation kernel and the clear image, and then the point-by-point multiplication results of all partitions are added together to obtain the observation model corresponding to the normalized image. The partition consistency windows satisfy the condition that the sum of the values ​​of all partition consistency windows equals 1 and each value is greater than or equal to 0. An objective function is constructed, consisting of three terms: the first term is the squared L2 norm of the difference between the normalized image and the spatial variation convolution result; the second term is the product of the squared negative part of the difference between the backfill intention weight field and the reference envelope energy and the spatial variation kernel envelope energy within the diffraction-limited radius for each partition, and the integral result of all partitions is multiplied by a first weighting coefficient; the third term is the total variation of the clear image multiplied by a second weighting coefficient. Both the first and second weighting coefficients are preset parameters. The first weighting coefficient ranges from 0.3 to 0.7, preferably 0.5, and the second weighting coefficient ranges from 0.02 to 0.1, preferably 0.05.

[0034] Specifically, the first term of the objective function The calculation formula is as follows: Where I represents the normalized image, Indicates a partition-consistent window. Let P represent the spatially variable kernel corresponding to the sliding window, X represent the clear image to be restored, and ⊙ represent point-by-point multiplication. This indicates a convolution operation.

[0035] Specifically, the second and third terms of the objective function The calculation formula is as follows: ,in and These represent the first weighting coefficient and the second weighting coefficient, respectively. Indicates the diffraction-limiting radius. This indicates the weight field of the intention to fill in the blanks. Indicates the empty variable nucleus The envelope energy curve, Indicates the reference envelope energy curve. Describes the negative part of the function. This represents the total variation in sharpness of an image.

[0036] It should be noted that the partitioned consistency window divides the entire image into multiple overlapping local regions, each partition corresponding to a window function. The window function value is non-negative, and the sum of the values ​​of the window functions of all partitions at each pixel position is 1. Its function is to decompose spatial blurring into a weighted concatenation of multiple local spatially varied convolutions, which ensures the accuracy of local blur modeling and achieves energy conservation and smooth transition in global concatenation. The first term of the objective function is used to constrain the pixel-level consistency between the restored sharp image after spatially varied kernel blurring and partitioned window concatenation and the input normalized image, ensuring that the restoration result fits the observation data. The second term of the objective function represents the weighted penalty for the portion of the spatially varied kernel envelope energy that is lower than the reference envelope energy within the diffraction limit radius of each partition. The weight is determined by the backfill intention weight field, and the purpose is to pull the energy of the spatially varied kernel back to the core region, specifically eliminating the central hollow phenomenon. The third term of the objective function applies total variation regularization to the sharp image, which maintains the image edge sharpness while suppressing noise amplification and avoiding artifacts introduced by overfitting during the restoration process.

[0037] It should be noted that the reference envelope energy curve can be obtained through experimental calibration. This involves setting up an ideal imaging environment that eliminates interference factors such as turbulence and inherent aberrations of the optical system, ensuring that the optical system is in a stable working state without additional aberrations. A standard point light source is selected and placed at the center of the object plane of the optical system and at multiple uniformly distributed field-of-view positions to ensure coverage of the effective imaging range of the system. The standard point light source is then imaged using a laser optical system, and multiple clear image sequences of the point light source intensity are acquired. The envelope energy curve is calculated for each acquired point light source intensity image using the same method. The average value of the envelope energy curves obtained from the calculation of multiple point light source images is then taken to obtain the reference envelope energy curve.

[0038] In one embodiment of the present invention, the sharp image is iteratively updated by alternating minimization, including the following steps: First, the convolution result of each partition's consistent window and the spatial variation kernel after flipping is calculated. This convolution result is multiplied point by point by the result of the normalized image divided by the spatial variation convolution result plus the numerical stability constant. Then, the point-by-point multiplication results of all partitions are added together. Next, the sum of the convolution result of each partition's consistent window and the spatial variation kernel after flipping plus the numerical stability constant is calculated. The former is divided by the latter and then multiplied point by point by the current sharp image to obtain the updated intermediate sharp image. Subsequently, the sharp image that minimizes the sum of the squared L2 variation regularization term of the difference between the intermediate sharp image and the sharp image to be updated is selected as the new sharp image. The spatially variable kernel is iteratively updated by alternating minimization, including the following steps: First, the residual is calculated, which is the difference between the normalized image and the spatially variable convolution result; then, for each partition, the pointwise product of the partition consistency window and the residual is calculated, and the convolution result after flipping the clear image is multiplied by negative two, plus the product of the weight coefficient and the gradient of the envelope energy penalty term of the spatially variable kernel. This result is multiplied by the step size and subtracted from the current spatially variable kernel to obtain an intermediate spatially variable kernel; finally, the intermediate spatially variable kernel is projected onto the feasible set of spatially variable kernels. This projection sequentially implements non-negativity constraints, unit sum constraints, envelope energy monotonicity constraints based on monotonic regression, and small energy threshold constraints at the diffraction limit radius to obtain a new spatially variable kernel.

[0039] It should be noted that the numerical stability constant is a preset parameter, with a value ranging from 10 to the power of -8 to -4. Preferably, the numerical stability constant is set to 10 to the power of -5. Its function is to avoid numerical overflow or instability caused by the denominator approaching zero during iteration, ensuring the robustness of the calculation process. Alternatingly minimizing and updating the clear image fixes the spatial variation kernel in each iteration. By optimizing the clear image, it is made to fit the input normalized image, and noise is suppressed and edges are kept clear through total variation regularization. The final clear image can accurately reproduce the details of the real scene, while having a low noise level and clearly distinguishable edge contours, avoiding artifacts introduced by overfitting the observation data. Alternatingly minimizing and updating the spatial variation kernel fixes the clear image in each iteration. The numerical distribution of the spatial variation kernel is optimized through gradient update and projected onto a feasible set of non-negative, unit sum, monotonically non-decreasing envelope energy, and core energy baseline, thereby accurately modeling the local spatial blurring characteristics.

[0040] In one embodiment of the present invention, controlling the iteration based on the convergence amount includes the following steps: Step S201: Using the restored image as the calculation object, calculate the envelope energy curve according to the process of central hollowness index measurement and spatial variation mapping, determine the minimum radius that reaches the small energy threshold, divide the minimum radius by the diffraction limit radius to obtain the instantaneous central hollowness index, take the median value of the instantaneous central hollowness index in the time dimension, and obtain the restored central hollowness index heat map. Step S202: Calculate the convergence factor, which is the median of the absolute values ​​of the differences between the hollowness index of the restoration center of all sliding windows and 1; calculate the data consistency residual, which is the squared L2 of the difference between the normalized image and the spatial variation convolution result. Step S203: Set the monitoring interval. At each iteration interval, calculate the absolute value of the difference between the current convergence and the convergence at the previous monitoring time, as well as the absolute value of the difference between the current data consistency residual and the data consistency residual at the previous monitoring time. Step S204: Determine whether to stop the iteration. When the current convergence is less than or equal to the convergence at the previous monitoring time, the absolute value of the difference between the convergences is less than or equal to the numerical tolerance, and the current data consistency residual is less than or equal to the data consistency residual at the previous monitoring time, the absolute value of the difference between the data consistency residuals is less than or equal to the numerical tolerance, the iteration is stopped. Step S205: If there are sliding windows with a hollowness index of the restoration center greater than 1, update the backfilling intention weight field corresponding to these sliding windows. The update method is to multiply the original backfilling intention weight field by 1 and add the product of the gain coefficient and the hollowness index of the restoration center minus 1. The value of the gain coefficient is greater than 0 and less than or equal to 1. Step S206: Repeat steps S201 to S205 until the stopping iteration condition is met.

[0041] It should be noted that the calculation of the hollowness index heatmap of the restored center, using the restored image as the object, requantifies the degree of core hollowness, providing a basis for iteration stopping and feedback; the calculation of convergence quantity and data consistency residual defines two quantifiable iterative monitoring quantities to improve the convergence accuracy, reflecting whether the core hollowness has been eliminated (convergence quantity) and whether the restoration result matches the input (data consistency residual), respectively; the frequency of iterative monitoring is set, and the convergence trend is judged by the difference of the index at intervals, balancing monitoring accuracy and computational efficiency, avoiding the computational overhead caused by monitoring in each iteration, where the monitoring interval is a preset parameter, preferably set to 8 times; the numerical tolerance is a preset parameter, preferably set to 5 × 10⁻⁴; the gain coefficient is a preset parameter, preferably set to 0.5, the gain coefficient is used to balance the need for feedback intensity and avoid over-correction; finally, automated iterative convergence is achieved without manual intervention, ensuring the reproducibility and engineering practicality of the process.

[0042] In one embodiment of the present invention, energy conservation verification is performed. The sum of the intensity values corresponding to all pixel coordinates and all frame numbers of the input image is calculated to obtain the total input intensity; the sum of the intensity values corresponding to all pixel coordinates and all frame numbers of the restored image is calculated to obtain the total output intensity; the absolute value of the difference between the total input intensity and the total output intensity is calculated as the energy difference. Envelope energy consistency verification is performed. For the spatially variant kernel of each sliding small window, with the center of the sliding small window as the center of the circle, for different radius ranges, the sum of the spatially variant kernel values corresponding to all pixel coordinates within each radius range is calculated respectively to form an envelope energy curve; at the diffraction limit radius, it is checked whether the value of the envelope energy curve is not less than the small energy threshold. Restored center hollowness index consistency verification is performed. Taking the restored image as the calculation object, according to the process of center hollowness index measurement and spatially variant mapping, the envelope energy curve is calculated, the minimum radius reaching the small energy threshold is determined, and the instantaneous center hollowness index is obtained by dividing the minimum radius by the diffraction limit radius. The median value of the instantaneous center hollowness index in the time dimension is taken to obtain the restored center hollowness index heat map; the median value of the absolute value of the difference between the restored center hollowness index of all sliding small windows and 1 is calculated.

[0043] It should be noted that an energy tolerance is set. When the energy difference is less than or equal to the energy tolerance, the energy conservation verification is determined to be qualified; if it is greater than the energy tolerance, it is determined to be unqualified; the number of small windows in which the value of the envelope energy curve at the diffraction limit radius is not less than the small energy threshold among all sliding small windows is counted, and the proportion of this number to the total number of sliding small windows is calculated. When this proportion is not lower than the set proportion threshold, the envelope energy consistency verification is determined to be qualified; otherwise, it is determined to be unqualified; a center hollowness index tolerance is set. When the median value of the absolute value of the difference between the restored center hollowness index of all sliding small windows and 1 is less than or equal to the tolerance, the restored center hollowness index consistency verification is determined to be qualified; otherwise, it is determined to be unqualified; the energy tolerance is a preset parameter, and its value range is from 10 to the power of negative 5 to 10 to the power of negative 3. Preferably, the energy tolerance is set to 5 times 10 to the power of negative 4; the proportion threshold of envelope energy consistency is a preset parameter, and its value range is from 90% to 100%. Preferably, the proportion threshold of envelope energy consistency is set to 98%; the center hollowness index tolerance is a preset parameter, and its value range is from 10 to the power of negative 3 to 10 to the power of negative 2. Preferably, the center hollowness index tolerance is set to 5 times 10 to the power of negative 3.

[0044] It should be noted that the energy conservation check ensures that no energy increases or decreases without cause during the restoration process by quantifying the difference in total intensity between the input and the restored image. The envelope energy consistency check ensures that all vacuous nuclei meet the core energy baseline requirements by examining the energy accumulation of the vacuous nuclei at the diffraction limit radius, thus avoiding restoration artifacts caused by insufficient local core energy. The restoration center hollowness index consistency check directly verifies whether the core objectives of the previous iteration have been achieved by quantifying the core hollowness of the restored image, ensuring that the energy distribution in the core region of the restored image meets expectations.

[0045] In one embodiment of the present invention, such as Figure 2 As shown, an artificial intelligence-based laser spatially variable PSF aberration elimination method includes the following steps: Step S301: Perform sliding window processing on the normalized image sequence, calculate the intensity centroid for each sliding window, and obtain the envelope energy curve with the intensity centroid as the center. Step S302: Determine the minimum radius through the small energy threshold, determine the central voidness index by combining the diffraction-limited radius, and construct a central voidness index heat map accordingly. Step S303: Establish a radial window smaller than the diffraction limit radius and map the central hollowness index to a backfill intention weight field; Step S304: Define a feasible set of empty variable kernels for each sliding window, and apply constraints that are non-negative, unity sum, monotonically non-decreasing envelope energy, and not lower than the small energy threshold at the diffraction limit radius; Step S305: A spatial variation convolutional observation model is established using a partitioned consistent window. An objective function is constructed that includes a data consistency term, a central envelope energy backfill term, and a total variation regularization term. The backfill intention weight field is used to weight the function and penalize insufficient envelope energy only within the range of less than the diffraction limit radius. Step S306: Iteratively update the clear image and the spatially variable kernel by alternating minimization, and project the image to the feasible set of the spatially variable kernel after each update. Step S307: Recalculate the heat map of the central hollowness index based on the restored image, determine the convergence amount accordingly, and control the iteration to stop or increase the backfill intention weight field at the corresponding position based on the convergence amount. Step S308: Output the restored image and the spatial variation kernel, and perform energy conservation check and envelope energy consistency check.

[0046] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0047] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A laser spatially variable PSF aberration elimination system based on artificial intelligence, characterized in that, include: The first module is used to perform sliding window processing on the normalized image sequence, calculate the intensity centroid for each sliding window, and obtain the envelope energy curve with the intensity centroid as the center. The second module is used to determine the minimum radius through a small energy threshold, combine it with the diffraction-limited radius to determine the central void index, and construct a central void index heat map accordingly. The third module is used to establish a radial window smaller than the diffraction limit radius and to map the central void index into a backfill intention weight field. The fourth module is used to define a feasible set of space-variable kernels for each sliding window, and to impose constraints that are non-negative, unity sum, monotonically non-decreasing envelope energy, and not lower than a small energy threshold at the diffraction-limited radius. The fifth module is used to establish a spatial variation convolutional observation model using a partitioned consistent window. It constructs an objective function that includes a data consistency term, a central envelope energy backfill term, and a total variation regularization term. The backfill intention weight field is used to weight the function and penalize insufficient envelope energy only within the range of the diffraction limit radius. The sixth module is used to iteratively update the clear image and the spatially variable kernel by alternating minimization, and to project to the feasible set of spatially variable kernels after each update; The seventh module is used to recalculate the heat map of the central hollowness index based on the restored image, determine the convergence amount accordingly, and control the iteration to stop or increase the backfill intention weight field at the corresponding position based on the convergence amount. The eighth module is used to output the restored image and the spatial variation kernel, and to perform energy conservation checks and envelope energy consistency checks.

2. The laser spatial variation PSF aberration elimination system based on artificial intelligence according to claim 1, characterized in that, For each frame of the image, energy normalization is performed within the region of interest so that the sum of the intensity values ​​of all pixels within that region of interest equals 1, thus obtaining a normalized image. Set up a sliding window with a binary mask. The pixel values ​​of the binary mask are only 0 or 1. Calculate the sum of the intensity values ​​of all pixels covered by the binary mask of the sliding window to obtain the total intensity of the window. Divide the intensity value of each pixel covered by the binary mask of the sliding window by the total intensity of the window to obtain the normalized intensity of the window for each corresponding pixel.

3. The laser spatial variation PSF aberration elimination system based on artificial intelligence according to claim 2, characterized in that, Using the normalized intensity of the small window of each pixel as the weight, the coordinates of all pixels covered by the binary mask of the sliding small window are summed in a weighted manner. Then, the weighted sum is divided by the sum of the normalized intensities of the small windows of all pixels to obtain the intensity centroid. Using the intensity centroid as the center, the sum of the normalized intensities of all pixels within each radius is calculated to form an envelope energy curve. A small energy threshold with a value greater than 0 and less than 1 is set, and the minimum radius that makes the envelope energy curve value reach the value of the small energy threshold is found. A diffraction-limited radius is set, and the found minimum radius is divided by the diffraction-limited radius to obtain the center hollowness index. The median of the center hollowness indices obtained for the same sliding window in all frame images is taken, and the median values ​​corresponding to all sliding windows together constitute the center hollowness index heatmap.

4. The laser spatial variation PSF aberration elimination system based on artificial intelligence according to claim 1, characterized in that, Extract the value corresponding to each sliding window in the heatmap of the central hollowness index, subtract 1 from the value to obtain the difference for each sliding window; multiply the difference for each sliding window by the scheduling intensity coefficient to obtain the first product for each sliding window; multiply the first product for each sliding window by the radial window value corresponding to different radial distance variables to obtain the second product for each sliding window at different radial distances; add 1 to the second product for each sliding window at different radial distances to obtain the backfilling intention weight for each sliding window at different radial distances. The backfilling intention weights of all sliding windows at different radial distances together constitute the backfilling intention weight field. radial window The calculation formula is as follows: Where r represents the radial distance variable, denoted by the diffraction-limiting radius, and exp denotes an exponential function with the natural constant as its base.

5. The laser spatially variable PSF aberration elimination system based on artificial intelligence according to claim 1, characterized in that, First, compare each value of the kernel to be projected with 0, take the larger value as the value of the intermediate kernel, calculate the sum of all the values ​​of the intermediate kernel, and divide each value of the intermediate kernel by the sum to obtain the normalized intermediate kernel. By defining a discrete sequence of radii, the sum of the values ​​of the normalized intermediate kernel within each discrete radius range is calculated to obtain the original cumulative energy sequence. For each discrete radius interval, calculate the scaling factor for that interval. This factor is the difference between the corresponding intervals in the new accumulated energy sequence and the sum of the values ​​of the normalized intermediate kernels in that interval. The normalized intermediate kernel is multiplied by the scaling factor of the corresponding interval for each discrete radius interval to obtain the monotonic kernel; Calculate the envelope energy of the monotonic nucleus at the diffraction-limited radius. If the value is less than the low energy threshold, calculate the gap value, which is the low energy threshold minus the envelope energy value. Define a core region and an outer region. The core region is the area with the center of the sliding window as the center and the diffraction limit radius as the radius. The outer region is the area outside the core region. Calculate the sum of the number of pixels in the core region and the number of pixels in the outer region. For the monotonic kernel values ​​within the core region, each value is increased by the gap value divided by the number of pixels in the core region; for the monotonic kernel values ​​within the outer region, each value is multiplied by 1 and subtracted from the gap value divided by the sum of the values ​​in the outer region to obtain the spatially variable kernel.

6. The laser spatially variable PSF aberration elimination system based on artificial intelligence according to claim 1, characterized in that, Each partition consistent window is multiplied pointwise by the convolution result of the corresponding spatial variable kernel and the clear image, and then the pointwise multiplication results of all partitions are added together to obtain the observation model corresponding to the normalized image. The partition consistent window satisfies that the sum of the values ​​of all partition consistent windows is equal to 1 and each value is greater than or equal to 0. The objective function is constructed, which consists of three terms: the first term is the squared L2 norm of the difference between the normalized image and the spatial variation convolution result; the second term is the product of the squared negative part of the difference between the backfill intention weight field and the reference envelope energy and the spatial variation kernel envelope energy within the diffraction limit radius for each partition, and then the integral result of all partitions is multiplied by the first weight coefficient; the third term is the total variation of the sharp image multiplied by the second weight coefficient.

7. The laser spatially variable PSF aberration elimination system based on artificial intelligence according to claim 1, characterized in that, The process of iteratively updating the sharp image using an alternating minimization method includes the following steps: First, calculate the convolution result of the consistent window of each partition and the spatial variation kernel after flipping. Multiply this convolution result point-by-point with the result of the normalized image divided by the spatial variation convolution result plus the numerical stability constant, and then add the point-by-point multiplication results of all partitions. Next, calculate the sum of the convolution result of the consistent window of each partition and the spatial variation kernel after flipping, plus the numerical stability constant. Divide the former by the latter, and then multiply it point-by-point with the current sharp image to obtain the updated intermediate sharp image. Finally, find the sharp image that minimizes the sum of the squared L2 norm of the difference between the intermediate sharp image and the sharp image to be updated, and use this as the new sharp image. The spatially variable kernel is iteratively updated by alternating minimization, including the following steps: First, the residual is calculated, which is the difference between the normalized image and the spatially variable convolution result; then, for each partition, the pointwise product of the partition consistency window and the residual is calculated, and the convolution result after flipping the clear image is multiplied by negative two, plus the product of the weight coefficient and the gradient of the envelope energy penalty term of the spatially variable kernel. This result is multiplied by the step size and subtracted from the current spatially variable kernel to obtain an intermediate spatially variable kernel; finally, the intermediate spatially variable kernel is projected onto the feasible set of spatially variable kernels. This projection sequentially implements non-negativity constraints, unit sum constraints, envelope energy monotonicity constraints based on monotonic regression, and small energy threshold constraints at the diffraction limit radius to obtain a new spatially variable kernel.

8. The laser spatially variable PSF aberration elimination system based on artificial intelligence according to claim 1, characterized in that, Iteration is controlled based on the convergence value, including the following steps: Step S201: Using the restored image as the calculation object, calculate the envelope energy curve according to the process of central hollowness index measurement and spatial variation mapping, determine the minimum radius that reaches the small energy threshold, divide the minimum radius by the diffraction limit radius to obtain the instantaneous central hollowness index, take the median value of the instantaneous central hollowness index in the time dimension, and obtain the restored central hollowness index heat map. Step S202: Calculate the convergence factor, which is the median of the absolute values ​​of the differences between the hollowness index of the restoration center of all sliding windows and 1; calculate the data consistency residual, which is the squared L2 of the difference between the normalized image and the spatial variation convolution result. Step S203: Set the monitoring interval. At each iteration interval, calculate the absolute value of the difference between the current convergence and the convergence at the previous monitoring time, as well as the absolute value of the difference between the current data consistency residual and the data consistency residual at the previous monitoring time. Step S204: Determine whether to stop the iteration. When the current convergence is less than or equal to the convergence at the previous monitoring time, the absolute value of the difference between the convergences is less than or equal to the numerical tolerance, and the current data consistency residual is less than or equal to the data consistency residual at the previous monitoring time, the absolute value of the difference between the data consistency residuals is less than or equal to the numerical tolerance, the iteration is stopped. Step S205: If there are sliding windows with a hollowness index of the restoration center greater than 1, update the backfilling intention weight field corresponding to these sliding windows. The update method is to multiply the original backfilling intention weight field by 1 and add the product of the gain coefficient and the hollowness index of the restoration center minus 1. The value of the gain coefficient is greater than 0 and less than or equal to 1. Step S206: Repeat steps S201 to S205 until the stopping iteration condition is met.

9. The laser spatially variable PSF aberration elimination system based on artificial intelligence according to claim 1, characterized in that, Perform energy conservation checks, calculate the sum of intensity values ​​corresponding to all pixel coordinates and all frame numbers of the input image to obtain the total input intensity; calculate the sum of intensity values ​​corresponding to all pixel coordinates and all frame numbers of the restored image to obtain the total output intensity. Calculate the absolute value of the difference between the total input intensity and the total output intensity, and use it as the energy difference; To perform an envelope energy consistency check, for each sliding window's spatially variable kernel, with the center of the sliding window as the center, calculate the sum of the spatially variable kernel values ​​corresponding to all pixel coordinates within each radius range to form an envelope energy curve; at the diffraction-limited radius, check whether the value of the envelope energy curve is not less than the low energy threshold. To verify the consistency of the restored center hollowness index, the restored image was used as the calculation object. Following the process of center hollowness index measurement and spatial variation mapping, the envelope energy curve was calculated to determine the minimum radius that reaches the small energy threshold. The minimum radius was divided by the diffraction limit radius to obtain the instantaneous center hollowness index. The median of the instantaneous center hollowness index in the time dimension was taken to obtain the heat map of the restored center hollowness index. The median of the absolute values ​​of the differences between the restored center hollowness index of all sliding windows and one was calculated.

10. A laser spatially variable PSF aberration elimination method based on artificial intelligence, characterized in that, Implementing an artificial intelligence-based laser spatially variable PSF aberration elimination system as described in claims 1 to 9 includes the following steps: Step S301: Perform sliding window processing on the normalized image sequence, calculate the intensity centroid for each sliding window, and obtain the envelope energy curve with the intensity centroid as the center. Step S302: Determine the minimum radius through the small energy threshold, determine the central voidness index by combining the diffraction-limited radius, and construct a central voidness index heat map accordingly. Step S303: Establish a radial window smaller than the diffraction limit radius and map the central hollowness index to a backfill intention weight field; Step S304: Define a feasible set of empty variable kernels for each sliding window, and apply constraints that are non-negative, unity sum, monotonically non-decreasing envelope energy, and not lower than the small energy threshold at the diffraction limit radius; Step S305: A spatial variation convolutional observation model is established using a partitioned consistent window. An objective function is constructed that includes a data consistency term, a central envelope energy backfill term, and a total variation regularization term. The backfill intention weight field is used to weight the function and penalize insufficient envelope energy only within the range of less than the diffraction limit radius. Step S306: Iteratively update the clear image and the spatially variable kernel by alternating minimization, and project the image to the feasible set of the spatially variable kernel after each update. Step S307: Recalculate the heat map of the central hollowness index based on the restored image, determine the convergence amount accordingly, and control the iteration to stop or increase the backfill intention weight field at the corresponding position based on the convergence amount. Step S308: Output the restored image and the spatial variation kernel, and perform energy conservation check and envelope energy consistency check.

Images