A point spread function optimization method for a multispectral compressive imaging system

Abstract

The application discloses a point spread function optimization method of a multispectral compressive imaging system, constructs two optimization objective functions of the multispectral compressive imaging system, and proposes a corresponding multi-objective optimization algorithm; one of the objective functions is used for evaluating light transfer performance, and the other is used for evaluating inter-correlation characteristics among multispectra; the optimization algorithm is based on the NSGAII algorithm, compared with a traditional genetic algorithm, the number of domination times is recorded, and the calculation complexity of the sorting process is effectively reduced; an elite strategy is introduced, a screening mechanism is introduced for individuals participating in the genetic, and the genetic process is optimized. The system point spread function after optimization can transfer more perfect spatial and spectral information to a greater extent, so that the subsequent reconstruction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of compressive imaging in computational optics, and in particular to a point spread function optimization method for a multi-spectral compressive imaging system. BACKGROUND

[0002] Spectral imaging can provide rich spatial and spectral information of an image, and is a new scientific and engineering application tool, which has wide application value in many fields. Traditional spectral imaging modes rely on spatial scanning or spectral scanning, and the scanning mechanism is relatively complex. For example, a swing scanning spectral imaging system images a row or a small area at a time, and spectral information is obtained through a light splitting component, and full field of view information is obtained through a scanning component and platform movement, and field of view splicing is required. The scanning spectral imaging can achieve high spatial and spectral resolution, but the time resolution is low, and is not suitable for scenes with high real-time requirements. Each sub-aperture of a multi-aperture spectral imaging system images a specific waveband alone, and the spatial and time resolutions are high, but the number of wavebands is limited.

[0003] In order to solve the problem of the trade-off between spatial resolution, time resolution and spectral resolution, a computational spectral imaging technology emerges as the times require. This technology uses a specially designed optical system to encode and compress the multi-spectral image information of a scene, and is received by a wide spectral range detector, and then a three-dimensional spectral image is reconstructed from a two-dimensional modulation image through an image reconstruction algorithm. In this way, the requirements for the imaging system hardware are more converted into the requirements for the algorithm, the light weight and miniaturization of the optical system can be realized, and the data amount of image storage and transmission is also reduced.

[0004] How to design a proper optical system to modulate the three-dimensional spatial-spectral information is one of the key technologies to realize computational spectral imaging. At present, multi-spectral information modulation is mainly realized through three ways, namely, image plane coding, spectral coding and point spread function engineering (Guo Jiaqi, Fan Benxuan, Liu Xin, et al. Computational spectral imaging: light field coding and algorithm decoding[J]. Laser and Optoelectronics Progress, 2024, 61(16).). Image plane coding is to image the object to be measured to the intermediate image plane, and to obtain the compressed image by compressing and coding the intermediate image. The coding function is generally a binary function, which removes redundant information by blocking pixels, multiplexes spectral information in space, and then images to the detector to reconstruct the measured image; the principle of spectral response coding method is to form a spectral filter array (SFA) through narrow-band filtering or specially designed multi-spectral response, and then to reconstruct the spectral image according to the complementary information provided by different channels. The key of this method lies in the design of spectral response function (SRF); the principle of point spread function (PSF) coding is to change the PSF of the system by using an additional optical element as a coding device, and to implement coding by using the characteristics of the wavelength change of PSF, and then to reconstruct the multi-spectral data cube. PSF coding is mainly based on wave optics, and generally uses diffractive optical elements (DOE) with certain microstructures to change the amplitude or phase distribution of incident light waves, and then to obtain the PSF that is conducive to spectral reconstruction.

[0005] Compared with image plane coding, which is only based on ideal geometric optics for analysis, it also needs to calibrate the reconstruction algorithm according to the aberration in actual application. The aberration caused by PSF in PSF coding, especially the chromatic aberration, can be directly used as the basis for spectral image reconstruction; compared with the spectral response method, although the SFA coding device is also very thin, its manufacturing difficulty is greater, and the design and manufacture of the PSF coding device DOE element are more mature and reliable. Therefore, the advantage of PSF coding is prominent, and it does not need to realize the explicit coding of the optical element, the optical system is simple and compact, and it can realize the miniaturization and lightness of the system, and even can directly transform the existing daily imaging equipment into a spectral imaging equipment. The present application is a related technical research on PSF coding.

[0006] For multispectral compressive imaging system, the quality of reconstructed image directly affects the application effect, but there is still a gap between the reconstruction effect and the quality of direct acquisition, which has become a key to restrict its further application. The quality of reconstructed image is closely related to the measurement matrix in the process of compressive imaging. A good measurement matrix can transfer more information flux in the sampling and reconstruction process, thus having higher quality of reconstructed image. The measurement matrix corresponds to the structure design of the encoding device in PSF encoding. At present, it is mainly randomly generated based on the theory of compressed sensing (CS) or artificially designed. The measurement matrix can be divided into deterministic measurement matrix and random measurement matrix (Tonggi. Research on Key Technologies of Infrared Compressive Imaging[D]. Harbin Engineering University[2023-11-06].). The deterministic measurement matrix generation methods mainly include polynomial measurement matrix, block polynomial measurement matrix, algebraic curve measurement matrix, etc. Its advantages are that it can save storage space through design, and it can also design the corresponding reconstruction algorithm to improve the speed and accuracy of reconstruction; the disadvantage is that it can only be designed for some specific scenarios, and it has high requirements for the sparsity of the signal. It is extremely difficult to design for high complexity signals such as natural scenes, and the corresponding reconstruction success rate is also very low. In contrast, the advantage of random measurement matrix is that it has low requirements for the sparsity of the target signal and is more adaptable to various natural scenes. Usually, the directly constructed measurement matrix is not the optimal one, so it is necessary to optimize the random measurement matrix and thus realize the optimization of PSF to improve the reconstruction quality. This optimization process is also called PSF engineering (Jia S, Vaughan J C, Zhuang X. Isotropic 3D Super-resolution Imaging with a Self-bending Point Spread Function[J]. Nature Photonics, 2014, 8(4): 302-306. DOI: 10.1038 / nphoton.2014.13.).

[0007] The current PSF engineering commonly used optimization algorithm mainly has genetic algorithm, ant colony algorithm and the like (Shechtman Y, Sahl SJ, Backer AS, et al. Optimal point spread function design for 3D imaging. [J]. Physical Review Letters, 2014, 113(13): 133902.), (Conkey D B, Brown A N, Caravaca-Aguirre A M, et al. Genetic algorithm optimization for focusing through turbid media in noisy environments [J]. Optics express, 2012, 20(5): 4840-4849.), and the optimization target needs to be designed according to different application scenarios. The objective function is usually optimized for the uniformity, definition and other indicators of the light spot (Feng, Fan, Yang, et al. Multi-objective optimization genetic algorithm for multi-point light focusing in wavefront shaping. [J]. Optics express, 2019, 27(25): 36459-36473.), and for the application field of the measurement matrix optimization of the multi-spectral compression imaging system, the optimization target and the optimization algorithm of the PSF engineering need to be reasonably designed according to the application requirement. The present application focuses on the research on the optimization design method of the PSF engineering of the multi-spectral compression imaging. SUMMARY

[0008] In order to solve the problems existing in the prior art, the present application aims to provide a point spread function optimization method of a multi-spectral compression imaging system, that is, the optimization design of the compression sampling matrix is optimized by an optimization algorithm, the optimization of the system PSF is realized, the optical transfer information flux is maximized and the correlation between the multi-spectrum is minimized, the imaging space resolution and the spectral resolution are increased, and the reconstruction precision is improved.

[0009] In order to achieve the purpose of the present application, the following technical solutions are adopted:

[0010] A point spread function optimization method of a multi-spectral compression imaging system, comprising the following steps:

[0011] (1) generating N initial random masks as an initial population, calculating the objective function (f1, f2) of each mask; wherein the first index f1 is the double integral of the modulation transfer function in the frequency domain space, and the second index f2 is the condition number of the inter-spectral point spread function correlation matrix;

[0012] (2) determining the hierarchy and the crowding distance according to the optimization objective function value (f1, f2) of each mask according to the non-dominated relationship, performing fast non-dominated sorting; generating four child phase templates; repeating the operations of selection, crossover and mutation N / 4 times to obtain the first generation of N child phase templates, and calculating the objective function (f1, f2) of each mask;

[0013] (3) mixing the N parent phase templates and the N child phase templates to obtain 2N phase templates; performing fast non-dominated sorting and crowding distance calculation according to the optimization objective function value (f1, f2) of each mask; retaining the first N child masks with lower hierarchical sequence number and larger crowding distance from the 2N mixed masks as new child masks to obtain a new population;

[0014] (4) judging whether the current population reaches the cycle number; if yes, outputting the N masks of the current population as the optimal solution; otherwise, returning to step (2) to repeat steps (2)-(3) on the new population; finally outputting the result as the best population, and the first one in the order is the best phase template.

[0015] Further, in the step (1), the f1 calculation formula is:

[0016]

[0017] wherein λ mid is the median spectrum in the multi-spectral band, and (u, v) is the frequency domain coordinate; f1 is used to evaluate the light transfer performance.

[0018] Further, in the step (1), the f2 calculation formula is:

[0019] f2=Cond(Corr),

[0020]

[0021]

[0022] wherein Cond represents the condition number of a matrix, and Corr is a matrix composed of the normalized cross-correlation matrix between the PSFs of each spectrum; the symbol ☆ represents cross-correlation operation; f2 is used to evaluate the cross-correlation characteristics between the multi-spectra.

[0023] Further, the objective function (f1, f2) formula is:

[0024] min{f1,f2}.

[0025] Further, in the step (2), the operations of selection, crossover and mutation are specifically:

[0026] Selection: In the current population P, according to the principle of minimum level of non-dominated sorting optimal front and minimum intra-layer congestion distance, according to the sorting result, four phase templates mask are selected as a group in turn, and the mask pair of two groups of parent phase templates is determined according to the four-element tournament principle;

[0027] Crossover: Each pair of parent phase templates is randomly and uniformly crossed according to the probability of 0-1;

[0028] Mutation: All masks are randomly inverted according to the probability of 0-1.

[0029] Further, the point spread function of the multi-spectral compression imaging system is optimized by using the NSGAII algorithm, wherein the main framework adopts the GA genetic algorithm.

[0030] Further, each phase template mask i corresponds to an individual element x i In order to make each objective function have equivalent influence on the congestion distance, f1 and f2 are normalized in the calculation process.

[0031] Further, the types of the phase templates include super-pixel random templates and Gaussian noise templates; and the corresponding optimization parameters are super-pixel template pixel values and Gaussian noise template Gaussian mean variance.

[0032] Further, the multi-spectral compression imaging system uses a parallel light pipe as a short-wave infrared target source, replaces a reticle and a wave plate as different target objects, and focuses the emitted parallel light on the object plane of a first diaphragm through a first imaging lens group with a focal length f to form an object plane target source, enters a 4f system, first passes through a polarizing mirror and a filter wheel for polarization and filtering in the 4f system to obtain polarized light of a selected waveband, a collimating lens is used as the first lens of the 4f system to realize Fourier frequency domain transformation of the target, the modulation in the frequency domain is realized by a random phase on a spatial light modulator through a beam splitter, a pair of orthogonal cylindrical lens groups are used as the second lens of the 4f system to realize Fourier inverse transformation, and finally the obtained modulation image is captured by a camera at the image plane.

[0033] Further, the multi-spectral compression imaging process of the multi-spectral compression imaging system includes point spread function calibration and multi-spectral imaging.

[0034] In the point spread function calibration process, the camera is set at the midpoint of the bifocal point of the cylindrical lens to obtain the best point spread function speckle; the regular speckles are disturbed by random modulation of the flat phase of the spatial light modulator on the Fourier plane, and the fine structure generated is regarded as a smooth speckle pattern; by controlling the light filter wheel, the point spread function speckles in the six spectrums are recorded in turn, and the point spread function calibration is completed;

[0035] In the multi-spectral imaging process, a real object is used instead of a point source to realize multi-spectral compressed imaging; when the corrected multi-spectral point spread function and the compressed image are available, the multi-spectral image is reconstructed by solving the ill-posed inverse problem with sparsity constraint.

[0036] The beneficial effects of the present application are as follows:

[0037] 1、The present application constructs two key indicators of the multi-spectral compressed imaging system, one evaluates the light transfer performance, and the other evaluates the cross-correlation characteristics between multi-spectrums. The designed optimization target can effectively increase the spatial and spectral information flux of the system, which helps to improve the reconstruction accuracy.

[0038] 2、The present application optimizes the PSF optimization performance by using NSGAII algorithm for multi-objective optimization problem. In the process of non-dominated sorting, the number of domination is recorded, and the crowding distance is introduced to increase the priority within the layer, further improve the performance of the optimization non-dominated sorting step, and effectively reduce the calculation complexity; the elite strategy is introduced, and the screening mechanism is introduced for the individuals participating in the genetic, to optimize the genetic process. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 It is a single-exposure multi-spectral compressed imaging system working principle schematic diagram of the present application;

[0040] Figure 2 It is a random phase mask distribution schematic diagram (superpixel size 60) of the present application;

[0041] Figure 3 It is a calibrated point spread function schematic diagram in the working spectral range of 1510nm-1560nm of the present application;

[0042] Figure 4 It is a multi-spectral compressed imaging system PSF optimization NSGAII algorithm flow chart of the present application. DETAILED DESCRIPTION

[0043] In order to better understand the technical scheme of the present application, the embodiments of the present application are described in detail below with reference to the drawings.

[0044] It should be apparent that the described embodiments are merely a few of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0045] The terms used in the embodiments of the present application are merely for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the embodiments of the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0046] The multi-spectral imaging optical device used in the embodiments of the present application is shown in FIG. 1. Figure 1 As shown in FIG. 1, a parallel light tube is used as a short-wave infrared (SWIR) target source, and by replacing the reticle and the wave plate as different target objects, the emitted parallel light is focused on the stop 1 (object plane) by the first imaging lens group with a focal length f = 200 mm to form an object plane target source, enters the 4f system (a typical filtering system in the optical field, which consists of two Fourier lens groups with a focal length f, and the distances from the object plane to the first lens, from the first lens to the second lens, and from the second lens to the image plane are single focal length f, double focal length 2f, and single focal length f, respectively), and the polarized light of the selected waveband is obtained by passing through the polarizer and the filter wheel in the 4f system first. The collimating lens serves as the first lens group of the 4f system, realizes the Fourier frequency domain transformation of the target, and the modulation in the frequency domain is realized by the random phase on the spatial light modulator (SLM) through the beam splitter. A pair of orthogonal cylindrical lens groups serves as the second lens of the 4f system, realizes the inverse Fourier transformation, and finally the obtained modulated image is captured by the SWIR camera at the image plane. In this process, the different spectral bands are ensured to be distinguishable by the astigmatism of the cylindrical lens. The specific multi-spectral compressed imaging process includes two steps: PSF calibration and multi-spectral imaging. In the PSF calibration step, the camera is set at the midpoint of the bifocal point of the cylindrical lens to obtain the best PSF speckle. The regular speckle is disturbed by the random modulation of the flat phase of the SLM on the Fourier plane, and a large number of fine structures are generated, which can be regarded as a smooth speckle pattern. By controlling the filter wheel, the PSF speckles in the six spectra are recorded in turn, and the PSF calibration is completed. In the multi-spectral imaging step, the real object is used to replace the point source to realize the multi-spectral compressed imaging. When the corrected multi-spectral PSF and the compressed image are available, the multi-spectral image can be reconstructed by solving the ill-posed inverse problem with sparsity constraint.

[0047] The PSF optimization process described in the present application is performed before the PSF calibration step, and is specifically designed as follows:

[0048] 1. First, the optimization target of PSF optimization design is selected.

[0049] On the one hand, it is desired that the PSF obtained through the measurement matrix coding realizes good general image quality and preserves as much detail information as possible, and the image variance, image information entropy, optical transfer function and the like can be used as indexes, and here the optical transfer function is selected to measure the information flux of the imaging system. Specifically, the modulus of the optical transfer function, i.e. the modulation transfer function (MTF), is calculated by double integration in the frequency domain space as an evaluation index of the system information flux:

[0050]

[0051] wherein λ mid is a median spectral band in the multi-spectral band, (u, v) is a frequency domain coordinate, and a negative value is taken to ensure that the optimization target is a minimum value. In order to reduce the influence of high-frequency noise in the frequency domain space, the present application only selects the central region as the integration range instead of the entire frequency domain space.

[0052] On the other hand, for multi-spectral compressed imaging, it is desired that the PSF between the spectral bands has a large degree of distinction under the premise of considering the imaging quality, so as to help improve the spectral resolution of the reconstructed image. This feature can be quantitatively measured by using the condition number of the cross-correlation coefficient matrix of the PSF between the spectral bands, and the formula is as follows:

[0053] f2 = Cond (Corr), (2)

[0054] wherein Cond represents the condition number of a matrix, and Corr is a matrix composed of the cross-correlation coefficients between the PSFs of each spectral band, and each matrix element is calculated according to the following steps: first, the cross-correlation coefficient between each PSF is calculated:

[0055]

[0056] The symbol ☆ in the formula represents the cross-correlation operation. In order to eliminate the difference in the absolute amplitude between each coefficient, the diagonal elements are normalized to obtain:

[0057]

[0058] The smaller the elements of the correlation matrix, the smaller the condition number, which means that the difference between the related spectral bands is greater, and the PSF is better. In particular, when the elements of the correlation matrix are zero, the correlation matrix is a diagonal matrix, and the condition number is the smallest, i.e. 1, at this time, each spectral band is not related, i.e. the difference between the spectral bands is the largest.

[0059] Therefore, the optimization target formula can be represented as formula (5):

[0060] min{f1,f2} (5)

[0061] 2、Secondly, the optimization algorithm of PSF optimization design is selected.

[0062] According to the foregoing, the number of optimization objectives is two, and therefore the optimization design problem of the PSF belongs to a "multi-objective nonlinear integer programming problem", and there are many optional optimization algorithms. Existing related researches on the wavefront shaping problem show that the optimization effect of a genetic algorithm is good, and researchers have made corresponding improvements and implementations for different specific applications. For the optimization problem with "multi-objective" optimization objectives, NSGAII (non-dominated sorting genetic algorithm II) is one of the most popular genetic algorithms, which is also the algorithm to be used in the present application (Deb K, Pratap A, Agarwal S, et al. A fast and elitist multiobjective genetic algorithm: NSGA-II, IEEE Trans. on Evol [J]. IEEE Transactions on Evolutionary Computation, 2002, 6.).

[0063] The main framework of the algorithm follows the genetic algorithm (GA), and has three steps of selection, crossover and mutation. Meanwhile, there are three key improvements compared with the genetic algorithm: first, for the processing of multiple optimization objectives, the non-dominated relationship is used to generate the Pareto front as the selection basis of the template individuals of the genetic algorithm; second, in the process of non-dominated sorting, the number of dominations is introduced for the rapid sorting of the template individuals, and the crowding distance is introduced for the priority sorting of the template individuals in the layer, and the combination of the two effectively reduces the computational complexity; third, the elite strategy is introduced, and a screening mechanism is introduced for the template individuals participating in the genetic, and the genetic process is optimized.

[0064] 3、In the optimization design process of the PSF, each phase mask mask i corresponds to an individual element x i In order to make each objective function have equivalent influence on the crowding distance, f1 and f2 should be normalized in the calculation process.

[0065] Finally, the type of the phase mask can be a superpixel random mask, a Gaussian noise mask and the like, and the optimization parameters can be the pixel value of the superpixel mask, the Gaussian mean variance of the Gaussian noise mask and the like.

[0066] After selecting the optimization objective and optimization algorithm, the specific implementation method of the PSF optimization design process is described as follows:

[0067] If a randomly distributed superpixel phase template is selected as the phase template type for modulation, then the optimization object is the pixel grayscale value of the template. Taking a 1920*1080 SLM scale as an example, a random phase template with a superpixel size of 60 is as follows: Figure 2 As shown. The system operates within a spectral range of 1510 nm to 1560 nm. A set of point spread functions is collected at 10 nm intervals, for a total of six point spread functions, as shown below. Figure 3 As shown. The specific algorithm flow for implementing PSF optimized template design using the NSGAII algorithm is as follows. Figure 4 As shown, the present invention provides a point spread function optimization method for a multispectral compressed imaging system, the detailed steps of which are as follows:

[0068] Step 1: Initial population generation.

[0069] N initial random masks are generated as the initial population P0, and the objective function (f1, f2) for each mask is calculated. The first index f1 is the double integral of the modulation transfer function (MTF) in the frequency domain space, calculated as shown in formula (1); the second index f2 is the condition number of the inter-spectral PSF correlation matrix, calculated as follows:

[0070]

[0071] f2 = Cond(Corr),

[0072]

[0073]

[0074] The optimization objective formula is as follows:

[0075] min{f1,f2},

[0076] Where, λ mid For the median spectral band in a multi-band matrix, (u,v) are frequency domain coordinates, and negative values ​​are used to ensure that the optimization objective is to take the minimum value; Cond represents the condition number of the matrix, and Corr is a matrix composed of normalized cross-correlation coefficients between the PSFs of each spectral band; the symbol ☆ represents cross-correlation operation.

[0077] Step 2: Genetically generate offspring individuals.

[0078] Based on the optimization objective function value (f1, f2) of each mask, the hierarchy and crowding distance are determined according to the non-dominated relationship, and a fast non-dominated sort is performed.

[0079] Selection: In the current population P, according to the principle of the minimum level of the non-dominated sorting optimal front and the minimum congestion distance in the layer, according to the sorting result, four phase masks are selected as a group in turn, combined in pairs according to the four-element tournament principle, and a pair of parent phase masks is determined.

[0080] Crossover: Each pair of parent phase masks is randomly and uniformly crossed according to a certain probability (0-1).

[0081] Mutation: All masks are randomly inverted according to a certain probability (0-1).

[0082] Thus, four child phase masks are generated. Repeat the above "selection, crossover, mutation" steps N / 4 times to obtain the first generation of N child phase masks, and calculate the objective function (f1, f2) of each mask.

[0083] Step 3: Elite reservation and new population determination.

[0084] Mix N parent phase masks with N child phase masks to obtain 2N phase masks.

[0085] According to the optimization objective function value (f1, f2) of each mask, perform fast non-dominated sorting and congestion distance calculation.

[0086] From the 2N mixed masks, the first N child masks with lower hierarchical sequence number and larger congestion distance are reserved as new child masks, where the hierarchical sequence number is given priority, and the congestion distance is considered in the case of the same sequence number. Thus, the new population is obtained.

[0087] Step 4: End condition judgment.

[0088] Determine whether the current population meets the cycle end condition of "reaching the cycle number". If it meets, output the N masks of the current population as the optimal solution; otherwise, return to step 2 and repeat steps 2-3 for the new population. The cycle number is taken into account comprehensively in terms of optimization effect and optimization time, and there is no fixed range. In this embodiment, the value is 20-100 times.

[0089] The final output result is the best population, and the first one in the ranking is the best phase mask. Thus, the PSF optimization is completed.

[0090] The phase mask obtained by the optimization result can be used in the subsequent multi-spectral compression imaging process, i.e. the process of "PSF calibration-imaging sampling-reconstruction". The optimized system PSF can transfer more perfect spatial and spectral information to a greater extent, thereby facilitating the improvement of the subsequent reconstruction accuracy.

[0091] The above merely describes the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-described embodiments. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall also be considered as falling within the protection scope of the present application.

Claims

1. A method of point spread function optimization for a multispectral compressive imaging system, comprising: It comprises the following steps: (1) generating N initial random masks as an initial population, calculating the objective function (f1, f2) of each mask; wherein the first index f1 is the double integral of the modulation transfer function in the frequency domain space, and the second index f2 is the condition number of the inter-spectral point spread function correlation matrix; (2) according to the optimization objective function value (f1, f2) of each mask, the hierarchy and the crowding distance are determined according to the non-dominated relationship, and the fast non-dominated sorting is carried out; four child phase templates are generated; the operations of selection, crossover and mutation are repeated N / 4 times to obtain N child phase templates of the first generation, and the objective function (f1, f2) of each mask is calculated; (3) mixing the N parent phase templates with the N child phase templates to obtain 2N phase templates; according to the optimization objective function value (f1, f2) of each mask, the fast non-dominated sorting and the crowding distance calculation are carried out; the first N child masks with lower hierarchical sequence number and larger crowding distance are reserved from the 2N mixed masks as new child masks to obtain a new population; (4) judging whether the current population reaches the cycle number; if yes, outputting the N masks of the current population as the optimal solution; otherwise, returning to step (2) to repeat steps (2)-(3) on the new population; finally outputting the result as the best population, and the first one in the sorting is the best phase template.

2. The method of claim 1, wherein, In the step (1), the f1 calculation formula is: where λ mid is the median spectral band of the multispectral band, (u, v) are the frequency domain coordinates; and f1is used to evaluate the light transfer performance.

3. The method of claim 2, wherein, In the step (1), the f2 calculation formula is: f2=Cond(Corr), Wherein, Cond represents the condition number of the matrix, Corr is a matrix composed of the normalized cross-correlation matrix between the PSFs of each spectral band; the symbol ☆ represents the cross-correlation operation; f2 is used to evaluate the cross-correlation characteristics between multiple spectrums.

4. The method of claim 3, wherein, The objective function (f1, f2) formula is: min{f1,f2}.

5. The method of claim 1, wherein, In the step (2), the selection, crossover and mutation operations are as follows: Selection: in the current population P, according to the principle of minimum hierarchy and minimum intra-layer crowding distance of the optimal front in the non-dominated sorting, according to the sorting result, four phase templates mask are selected as a group in turn, and the mask pair of two parent phase templates is determined according to the four-element tournament principle; Crossover: each pair of parent phase templates is randomly and uniformly crossed according to the probability of 0-1; Mutation: all masks are randomly inverted according to the probability of 0-1.

6. The method of claim 1, wherein, The NSGAII algorithm is used to optimize the point spread function of the multi-spectral compression imaging system, wherein the main framework adopts the GA genetic algorithm.

7. The method of claim 1, wherein, Each phase template mask i Corresponding to one body element x i In order to make each objective function have equivalent influence on the congestion distance, f1 and f2 are normalized respectively in the calculation process.

8. The method of claim 1, wherein, The types of the phase templates include super-pixel random templates and Gaussian noise templates; and the corresponding optimization parameters are the pixel values of the super-pixel templates and the Gaussian mean variance of the Gaussian noise templates.

9. The method of claim 1, wherein, The multispectral compressive imaging system uses a parallel light pipe as a short-wave infrared target source, replaces a reticle and a wave plate as different target objects, and focuses the emitted parallel light on an object plane of a first diaphragm by a first imaging lens group with a focal length f to form an object plane target source, enters a 4f system, and is first subjected to polarization and filtering by a polarizing mirror and a filter wheel in the 4f system to obtain polarized light of a selected waveband, collimates the light by a collimator lens as a first lens of the 4f system to realize Fourier frequency domain transformation of the target, is modulated in the frequency domain by a random phase on a spatial light modulator through a beam splitter, is subjected to Fourier inverse transformation by a pair of orthogonal cylindrical lens groups as a second lens of the 4f system, and is finally captured by a camera at an image plane.

10. The method of claim 9, wherein, The multispectral compressive imaging process of the multispectral compressive imaging system includes point spread function calibration and multispectral imaging. In the point spread function calibration process, the camera is arranged at a midpoint of a bifocal point of the cylindrical lens to obtain optimal point spread function speckles; the regular speckles are disturbed by random modulation of a flat phase of a spatial light modulator on a Fourier plane to generate fine structures regarded as smooth speckle patterns; and the point spread function speckles in six spectrums are recorded in sequence by controlling a filter wheel to complete the point spread function calibration. In the multispectral imaging process, a real object is used to replace a point source to realize multispectral compressive imaging; and when the calibrated multispectral point spread function and the compressed image are available, the multispectral image is reconstructed by solving an ill-posed inverse problem with a sparsity constraint.

Images