Diffraction and light field fused hyperspectral imaging system and method

Abstract

The invention relates to the technical field of computational optical imaging, in particular to a diffraction and light field fused hyperspectral imaging system and method. The system adopts a snapshot structure in which a diffraction optical channel and a panchromatic imaging channel are parallel; the diffractive optical channel acquires aliasing spectrum light field information including space and angle information through the axial dispersion characteristic of the diffractive primary mirror and the light field imaging component; and the panchromatic imaging channel synchronously acquires a high-spatial-resolution panchromatic image. The method comprises the following steps of: performing depth estimation on aliasing spectrum light field information by using a panchromatic image as spatial structure prior to obtain a scene depth map; determining the position of a refocusing plane according to the scene depth map and diffraction dispersion characteristics, carrying out digital refocusing, and generating a fuzzy spectrum slice; and finally, by taking the panchromatic image as a spatial constraint, decoupling and reconstructing a hyperspectral data cube through joint deconvolution operation. According to the invention, a mechanical scanning link is eliminated, and the system stability and the imaging efficiency are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of computational optical imaging technology, and in particular to a hyperspectral imaging system and method that fuses diffraction and light field. Background Technology

[0002] Existing hyperspectral imaging technologies that fuse diffraction and light fields suffer from the following technical challenges: existing diffraction spectral imaging methods rely on the mechanical displacement of detectors or optical elements along the optical axis to achieve full-spectrum scanning. In demanding applications such as space remote sensing and on-orbit scientific payloads, which are sensitive to vibration and difficult to maintain, this mechanical motion structure is prone to introducing attitude errors and reducing system reliability. At the same time, the point-by-point scanning mechanism leads to prolonged image acquisition time, making it impossible to monitor dynamic targets or transient phenomena in real time. For example, in Earth observation, it is difficult to quickly capture spectral information of cloud movement or ocean surface changes, thus limiting its application on lightweight platforms that require high timeliness and stability. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a hyperspectral imaging system and method that integrates diffraction and light field, solving the technical problems of low system stability, slow imaging speed, and difficulty in meeting the stringent requirements of lightweight platforms caused by the reliance on mechanical scanning components in existing diffraction spectral imaging technologies.

[0004] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows:

[0005] In a first aspect, the present invention provides a hyperspectral imaging system that fuses diffraction and light field, comprising:

[0006] The aliased spectral light field information of the target scene after dispersion by the diffraction primary mirror is acquired through the diffraction optical channel. The aliased spectral light field information includes the spatial and angular information of the light rays. At the same time, a high spatial resolution panchromatic image of the target scene is acquired through an independent panchromatic imaging channel.

[0007] Using the spatial structure prior provided by the panchromatic image, depth estimation is performed on the aliased spectral light field information to obtain a scene depth map;

[0008] Based on the scene depth map and diffraction dispersion characteristics, a series of refocusing plane positions corresponding to different center wavelengths are determined;

[0009] The aliased spectral light field information is digitally refocused at the refocusing plane position to generate a set of blurred spectral slices distributed along the optical axis.

[0010] Using the panchromatic image as a spatial constraint, a joint deconvolution operation is performed on the blurred spectral slices to decouple and reconstruct the hyperspectral data cube of the target scene.

[0011] Furthermore, in the hyperspectral imaging system for diffraction and light field fusion described in this invention, the step of using the spatial structure prior provided by the panchromatic image to perform depth estimation on the aliased spectral light field information to obtain a scene depth map includes:

[0012] The aliased spectral light field information is decoded to extract a set of sub-aperture images from multiple different perspectives;

[0013] Calculate the pixel displacement of the same spatial feature point in the sub-aperture image set between adjacent sub-aperture images to obtain an initial disparity map;

[0014] Convert the initial disparity map into an initial depth map;

[0015] The panchromatic image is registered with the central viewpoint image extracted from the aliased spectral light field information with high precision to establish a pixel-level correspondence.

[0016] Using the texture details of the registered panchromatic image as constraints, the initial depth map is corrected through an optimization algorithm to fill in the missing depth information caused by weak textures or occluded areas, smooth depth noise, and output a high-precision and spatially consistent scene depth map.

[0017] Furthermore, in the hyperspectral imaging system for diffraction and light field fusion described in this invention, the step of digitally refocusing the aliased spectral light field information at the refocusing plane position to generate a set of blurred spectral slices distributed along the optical axis includes:

[0018] Based on the dispersion equation of the diffraction primary mirror, the preset spectral band range is mapped to the optical axis direction, and the position of the refocusing plane corresponding to each spectral channel is calculated.

[0019] For each of the refocusing plane positions, the coordinate transformation of the four-dimensional light field data is performed according to the light field propagation model to simulate the process of light rays reconverging at the refocusing plane position;

[0020] The transformed light field data is integrated at the refocusing plane position to synthesize the two-dimensional projection image corresponding to that position, i.e., the blurred spectral slice;

[0021] In this process, on the refocusing plane corresponding to a specific wavelength, the light signal of that wavelength is in a quasi-focused state, while the light signals of other wavelengths are blurred due to defocusing, so that each of the blurred spectral slices is an aliasing of multi-band signals.

[0022] Furthermore, in the hyperspectral imaging system for diffraction and light field fusion described in this invention, the step of performing joint deconvolution operations on the blurred spectral slices using the panchromatic image as a spatial constraint to decouple and reconstruct the hyperspectral data cube of the target scene includes:

[0023] A linear model of system imaging is constructed, in which the blurred spectral slice is expressed as the convolution result of the hyperspectral data cube to be reconstructed and the three-dimensional point spread function that varies with depth;

[0024] Establish an inversion objective function that includes a data fidelity term and a regularization term;

[0025] The information of the panchromatic image is introduced into the regularization term as a spatial constraint, wherein the intensity information of the panchromatic image is used to constrain the sum of radiance of the reconstruction results of the hyperspectral data cube in each spectral band;

[0026] An iterative optimization algorithm is used to solve the inversion objective function. Under the spatial constraints of the panchromatic image, the spatial images of all spectral bands are simultaneously decoupled from the blurred spectral slices, and the hyperspectral data cube is finally output.

[0027] Furthermore, in the hyperspectral imaging system for diffraction and light field fusion described in this invention, the acquisition of the aliased spectral light field information of the target scene after dispersion by the diffraction primary mirror through the diffraction optical channel includes:

[0028] The incident light of the target scene passes through the diffraction mirror. According to its axial dispersion characteristics, the diffraction mirror disperses light components of different wavelengths along the optical axis and converges them at different positions on the optical axis.

[0029] Subsequently, the light carrying wavelength and spatial information enters the light field imaging component, which consists of a light field acquisition lens and a microlens array;

[0030] The microlens array separates light rays from different propagation directions and guides them to different pixel regions of the imaging detector, thereby simultaneously recording the spatial position and angular distribution of light rays in a single exposure and obtaining the aliased spectral light field information.

[0031] Furthermore, in the hyperspectral imaging system of diffraction and light field fusion described in this invention, the high spatial resolution panchromatic image of the target scene is acquired through an independent panchromatic imaging channel, wherein:

[0032] The optical path of the panchromatic imaging channel is set in parallel with the optical path of the diffractive optical channel, and they respectively image the same target scene.

[0033] The panchromatic imaging channel includes a front lens group for imaging the target scene onto a second imaging detector;

[0034] The panchromatic imaging channel does not include dispersive elements and microlens arrays, allowing the second imaging detector to directly receive the integrated intensity information of the target scene, thereby obtaining the panchromatic image with a spatial resolution higher than that of the diffraction optical channel.

[0035] In a second aspect, the present invention provides a hyperspectral imaging system for fusing diffraction and light field, for performing the aforementioned hyperspectral imaging system for fusing diffraction and light field, comprising:

[0036] The diffraction optical channel has a diffraction primary mirror, a light field acquisition lens, a microlens array and a first imaging detector arranged sequentially in its optical path to acquire aliased spectral light field information.

[0037] A panchromatic imaging channel has a front lens group and a second imaging detector arranged in its optical path to acquire high spatial resolution panchromatic images. The optical path of the panchromatic imaging channel is parallel to the optical path of the diffractive optical channel.

[0038] The processor is configured to: receive the aliased spectral light field information and the panchromatic image; use the panchromatic image to perform depth estimation on the aliased spectral light field information to obtain a scene depth map; perform digital refocusing based on the scene depth map to generate a blurred spectral slice; and use the panchromatic image as a spatial constraint to perform joint deconvolution operation on the blurred spectral slice to reconstruct a hyperspectral data cube.

[0039] Beneficial effects of this invention;

[0040] The hyperspectral imaging system and method provided by this invention, through a parallel snapshot-style acquisition design of the diffraction optical channel and the panchromatic imaging channel, fundamentally eliminates the dependence on mechanical scanning components. This significantly improves the structural reliability and stability of the system in demanding application scenarios such as space remote sensing and on-orbit scientific payloads, and enables real-time spectral monitoring of dynamic targets or transient phenomena. Furthermore, by utilizing the high spatial resolution structural prior information provided by the panchromatic image, depth estimation and correction are performed on the aliased spectral light field information originating from the diffraction-light field channel, effectively overcoming the shortcomings of existing light field depth estimation in weak texture or occluded regions. Obtaining a high-precision scene depth map lays a precise geometric foundation for subsequent spectral reconstruction. Based on this, digital refocusing combined with diffraction dispersion characteristics can accurately separate the blurred spectral slices distributed along the optical axis. Integrating the panchromatic image as a spatial constraint into the joint deconvolution operation on the blurred spectral slices constructs a strongly constrained inversion problem. This mechanism can effectively suppress aliasing and blurring components in the spectral decoupling process, thus reconstructing a hyperspectral data cube with both high spatial resolution and high spectral fidelity without the need for physical scanning. This successfully solves the inherent contradiction between volume, resolution, and imaging speed in existing technologies. Attached Figure Description

[0041] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the architecture of a hyperspectral imaging system that integrates diffraction and light field according to the present invention. Detailed Implementation

[0043] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0044] Please see Figure 1 This invention provides a hyperspectral imaging system that fuses diffraction and light field, comprising a diffraction primary mirror 1, a light field acquisition lens 2, a microlens array 3, and an imaging detector 4. The light field acquisition lens 2, microlens array 3, and imaging detector 4 constitute a light field sensor. Each microlens corresponds to multiple detector pixels. Through dispersion along the axis by the diffraction primary mirror and refraction by the light field lens, the spectral image of the scene can be focused in front of the microlens array. The microlenses then re-image this focused image and project it onto a high-resolution image sensor to obtain aliased spectral information.

[0045] The panchromatic imaging channel is used to acquire a reference image with high spatial resolution. Its optical path is parallel to that of the hyperspectral imaging channel and is formed by the front lens group 5 and the second imaging detector 6.

[0046] When the system is working, the incident light from the target scene enters two channels respectively; one enters the aforementioned diffraction-light field imaging channel, where the light field sensor captures the aliased spectral information; the other enters the panchromatic imaging channel, where the front lens group 5 and the second imaging detector 6 acquire a high-resolution panchromatic image of the target.

[0047] The diffraction primary mirror 1, serving as the system's front optical element, is responsible for receiving light radiation from the target scene. It operates based on the principle of diffraction rather than refraction, possessing unique axial dispersion characteristics; the incident wavelength is inversely proportional to its effective focal length.

[0048]

[0049] Where f0 is the focal length of the DOE design wavelength λ0, and D is the design constant.

[0050] The equation for a first-order diffractive lens is:

[0051]

[0052] As can be seen from this formula, the image distance z along the optical axis mainly depends on the wavelength A0.

[0053] Therefore, light of different wavelengths in the same target scene will form clear images at different positions along the optical axis after passing through the diffraction mirror 1, thus producing a dispersion and spectral layering imaging effect along the optical axis.

[0054] The light rays, after being diffracted and initially converged by the diffraction mirror 1, then enter the light field sensor, which consists of the light field acquisition lens 2, the microlens array 3, and the imaging detector 4.

[0055] In existing camera imaging processes, light rays emitted from an object point within a certain angular range from different directions converge at only one point on the imaging detector after passing through the main lens, resulting in the loss of directional information. However, a light field camera based on a microlens array couples a microlens array between the main lens and the imaging detector, allowing light rays from different directions to pass through the microlenses and then disperse to different positions on the image sensor after being focused by the light field acquisition lens.

[0056] This system is based on spatial multiplexing, simultaneously recording the spatial information (x, y coordinates) and angular information (u, v directions) of light in a single exposure, thus capturing four-dimensional light field data. The final image obtained on the target surface of imaging detector 4 is a mixed spectral information modulated by diffraction dispersion and microlens angle modulation.

[0057] The dual-channel data captured by the system (diffraction-light field aliasing image and panchromatic image) needs to undergo a series of advanced computational imaging processes to finally reconstruct a clear and accurate hyperspectral data cube. The core of this restoration process includes two interconnected and sequential parts: light field refocusing based on depth estimation and spectral decoupling based on three-dimensional deconvolution.

[0058] Light field refocusing based on depth estimation;

[0059] The purpose of this step is to use the angular information of the light field data to recover the three-dimensional geometry of the scene, and to initially separate the spectral images of different wavelengths distributed along the optical axis using digital refocusing technology.

[0060] The system first decodes the four-dimensional light field data acquired through the diffraction-light field channel to generate a series of sub-aperture images with slight parallax. By calculating the spatial displacement (i.e., parallax) of the same object point in each sub-aperture image and based on the principles of geometric optics, the depth information of that object point relative to the system can be calculated.

[0061] To improve the accuracy of depth estimation, especially in areas with weak texture or occlusion, this invention introduces a high-resolution image acquired through the panchromatic channel as a spatial prior. The panchromatic image contains rich details of scene structure. Through joint optimization algorithms (such as image registration, Markov random field optimization, etc.), the initial depth map is constrained and corrected, ultimately outputting a high-precision, spatially consistent scene depth map.

[0062] Based on the theory of light field propagation, by translating and integrating four-dimensional light field data, an image focused on any virtual plane can be digitally synthesized. According to the dispersion equation of the diffraction primary mirror (Equation (2)), the target spectral range [λ] is... min ,λ max Mapped to a series of specific refocusing plane positions λ on the optical axis k , where k = 1, 2, ..., N, and N is the number of spectral bands.

[0063] The high-precision depth map obtained is used to guide the refocusing process. The system performs digital refocusing along the optical axis z on planes corresponding to each center wavelength. This process generates a series of images I(x,y;λ). k Due to diffraction dispersion, at λ k On the plane, only the light signal of that wavelength is in focus, while all other light signals of different wavelengths are superimposed on it with varying degrees of blur. Therefore, each I(x,y;λ) frame... k All of these are the same goal in λ. k The aliasing of the focused image of the wavelength with the defocused and blurred images of all other bands at that location is a blurred spectral slice, which serves as input data for subsequent spectral decoupling.

[0064] Spectral decoupling based on 3D deconvolution;

[0065] From the aforementioned aliased and blurred slices, an independent and clear spatial image of each spectral band is precisely separated and recovered.

[0066] The imaging process of the system can be precisely modeled in the spatial domain as a three-dimensional convolution process:

[0067]

[0068] O(x, y, z) is the original three-dimensional hyperspectral data cube to be determined. h(x, y, Δz) is the system's three-dimensional point spread function (PSF) varying with the defocusing amount Δz, characterizing the blurring properties of the optical system. I(x i y i , z i ) is at position z i The blurred slice obtained from the location.

[0069] Discretizing the model and representing it in the frequency domain simplifies it to a linear equation, from which the spectral signal O can be obtained through inverse operations.

[0070] O = H -1 I

[0071] The full color information is S PAN Panchromatic information is the total information obtained by superimposing all spectral information, i.e., S. PAN =O1+O2+O3+O4, which can be used as additional information for joint calculation. A regularized deconvolution framework is adopted, incorporating full-color channel information as a key constraint into the calculation process.

[0072] Through the above computational processing based on light field refocusing and joint inversion spectral decoupling, the system finally outputs the restored hyperspectral data cube O(x,y,λ). k This data cube combines high spatial resolution with high spectral fidelity.

[0073] This invention provides a hyperspectral imaging system that fuses diffraction and light field, comprising:

[0074] The aliased spectral light field information of the target scene after dispersion by the diffraction primary mirror is acquired through the diffraction optical channel. The aliased spectral light field information includes the spatial and angular information of the light rays. At the same time, a high spatial resolution panchromatic image of the target scene is acquired through an independent panchromatic imaging channel.

[0075] Using the spatial structure prior provided by the panchromatic image, depth estimation is performed on the aliased spectral light field information to obtain a scene depth map;

[0076] Based on the scene depth map and diffraction dispersion characteristics, a series of refocusing plane positions corresponding to different center wavelengths are determined;

[0077] The aliased spectral light field information is digitally refocused at the refocusing plane position to generate a set of blurred spectral slices distributed along the optical axis.

[0078] Using the panchromatic image as a spatial constraint, a joint deconvolution operation is performed on the blurred spectral slices to decouple and reconstruct the hyperspectral data cube of the target scene.

[0079] Incident light from the target scene is first received by the diffraction primary mirror. Based on its unique axial dispersion characteristics, the diffraction primary mirror disperses light components of different wavelengths along the optical axis, causing the light rays of each wavelength to converge at different positions on the optical axis. Subsequently, this portion of the light carrying wavelength and spatial information enters the diffraction optical channel. The diffraction optical channel consists of a light field acquisition lens, a microlens array, and a first imaging detector. After being refracted by the light field acquisition lens, the light is guided to the microlens array. The microlens array separates the light rays from different propagation directions and images them onto different pixel areas of the first imaging detector. In this way, the system simultaneously records the spatial position and angular distribution of the light in a single exposure, thereby obtaining aliased spectral light field information including four-dimensional light field information. Simultaneously, another incident light enters a separately configured panchromatic imaging channel. The panchromatic imaging channel is set in parallel with the diffraction optical channel, and its optical path includes a front lens group and a second imaging detector. Since the panchromatic imaging channel does not contain any dispersive elements or microlens arrays, the second imaging detector directly receives the integrated intensity information of the target scene, thereby obtaining a high spatial resolution panchromatic image with a higher spatial resolution than the information obtained by the diffraction optical channel.

[0080] During the data computation phase, the processor receives the aforementioned dual-channel information. The depth estimation process begins with decoding the aliased spectral light field information. After decoding, multiple sets of sub-aperture images with small disparities are extracted. By calculating the pixel displacement of the same spatial feature point between adjacent images in the sub-aperture image set, an initial disparity map can be obtained. Subsequently, the initial disparity map is converted into an initial depth map. To improve the accuracy of depth estimation, especially to compensate for the lack of information in weak textures or occluded areas, a panchromatic image is introduced as a spatial structure prior. Specifically, the panchromatic image is first highly registered with the center viewpoint image extracted from the aliased spectral light field information to establish a pixel-level correspondence. Then, constrained by the rich texture details contained in the registered panchromatic image, a specific optimization algorithm is used to correct the initial depth map. This correction process can fill in the areas where depth information is missing and smooth the noise in the depth map, ultimately outputting a high-precision scene depth map with good spatial consistency.

[0081] Next, based on the obtained scene depth map and the dispersion characteristics of the diffraction primary mirror, the plane position requiring digital refocusing is determined. Specifically, according to the dispersion equation of the diffraction primary mirror, the target spectral band range is mapped onto the optical axis, and the refocusing plane position corresponding to the center wavelength of each spectral channel is calculated. The digital refocusing operation is performed for each refocusing plane position. For a specific position, according to the light field propagation model, the four-dimensional light field data is transformed to simulate the process of light reconverging at this plane position. Then, the transformed light field data is integrated at this refocusing plane position to synthesize the corresponding two-dimensional projected image, i.e., the blurred spectral slice. Due to the diffraction dispersion effect, on the refocusing plane corresponding to a specific wavelength, only the light signal of that wavelength is in quasi-focus, while all other wavelengths are in defocus and blurred. Therefore, each blurred spectral slice is essentially the result of multi-band signal aliasing.

[0082] The final stage involves spectral decoupling to reconstruct the hyperspectral data cube. This process uses a panchromatic image as a key spatial constraint. First, a linear model of the system imaging is constructed, representing the acquired blurred spectral slices as the result of convolving the hyperspectral data cube to be reconstructed with a depth-varying three-dimensional point spread function. Then, an inversion objective function is established, including data fidelity and regularization terms. Information from the panchromatic image is incorporated into the regularization term; the intensity information of the panchromatic image is used to constrain the sum of radiance of the reconstructed hyperspectral data cube across all spectral bands to be consistent with that of the panchromatic image. An iterative optimization algorithm is used to solve this inversion objective function. Under the spatial constraints provided by the panchromatic image, the optimization algorithm simultaneously decouples clear spatial images of all spectral bands from a set of blurred spectral slices, ultimately outputting a complete hyperspectral data cube. This technical solution achieves high-quality hyperspectral imaging without mechanical scanning through the coordinated and progressive processing of dual-channel information.

[0083] The specific operation of decoding the aliased spectral light field information involves reprojecting the light propagation direction by analyzing the dot matrix image formed by the microlens array on the imaging detector and, based on the correspondence between each microlens and its subordinate pixels. This decoding process extracts a set of sub-aperture images from multiple different perspectives, each representing a scene observed from a specific, small viewpoint. The pixel displacement of the same spatial feature point in the sub-aperture image set between adjacent sub-aperture images is calculated; this displacement, or parallax, is inversely proportional to the scene depth. This calculation yields the initial parallax map. The initial parallax map is converted into an initial depth map based on geometric optics principles, utilizing camera intrinsic and extrinsic parameters and the mathematical relationship between parallax and depth. High-precision registration of the panchromatic image with the central viewpoint image extracted from the aliased spectral light field information establishes a precise pixel-level correspondence between the two images, ensuring that the details of the panchromatic image are accurately mapped to the coordinate space of the depth map. Using the texture details of the registered panchromatic image as a constraint, the initial depth map is corrected through an optimization algorithm. This optimization algorithm utilizes the clear edge and texture information in the panchromatic image to guide the depth estimation to produce reasonable results in the weak texture region and suppress noise in the depth map, ultimately outputting a high-precision and spatially consistent scene depth map.

[0084] Based on the dispersion equation of the diffraction primary mirror, the preset spectral band range is mapped to the optical axis. This is done by calculating the position of the sharp imaging plane corresponding to the center wavelength of each spectral channel of interest to the spectral imaging system, based on the functional relationship between wavelength and image distance. Each position is the refocusing plane position. For each calculated refocusing plane position, the four-dimensional light field data is transformed according to the light field propagation model. This transformation essentially simulates how the light should be redistributed on the imaging detector when the virtual imaging plane is moved to that refocusing plane position. The transformed light field data is integrated at the refocusing plane position. This integration operation is equivalent to accumulating the light intensities from all directions reaching the virtual plane, thereby synthesizing the two-dimensional projected image corresponding to that position, which is the blurred spectral slice. On the refocusing plane corresponding to a specific wavelength, due to the dispersion characteristics of the diffraction primary mirror, only light near that wavelength can be clearly focused on this plane. Light of other wavelengths will form defocused blur spots on this plane. Therefore, each blurred spectral slice is a superposition of the sharp image of the target scene at that wavelength component and the blurred image of other wavelength components, forming a multi-band signal aliasing.

[0085] Constructing a linear model for system imaging approximates the physical imaging process of the optical system as a linear system. The blurred spectral slice is described as the result of convolving an ideal hyperspectral data cube with a three-dimensional point spread function characterizing the system's optical properties. An inversion objective function is established, including data fidelity and regularization terms. The data fidelity term requires the reconstructed result to be as consistent as possible with the observed blurred spectral slice, while the regularization term introduces prior knowledge about the solution to stabilize the inversion process and obtain the physical solution. Information from the panchromatic image is incorporated into the regularization term as a spatial constraint; specifically, the pixel intensity of the panchromatic image is used as a constraint condition because the intensity of the panchromatic image is approximately equal to the sum of the radiant intensities of all bands in the hyperspectral data cube at that location, providing a strong spatial constraint for spectral decoupling. An iterative optimization algorithm is used to solve the inversion objective function. Guided by the spatial constraints provided by the panchromatic image, the algorithm gradually adjusts the estimated value of the hyperspectral data cube so that the final solution not only conforms to the observed fuzzy spectral slice data, but also satisfies the spatial smoothness and total radiometric constraints specified by the panchromatic image. Thus, it simultaneously decouples clear spatial images of all spectral bands from the aliased fuzzy spectral slices.

[0086] The incident light from the target scene passes through a diffraction mirror. The mirror, based on its axial dispersion characteristics, modulates the light wave's phase using micro-nano structures, unlike the refraction principle of existing lenses. This causes different wavelengths of light to be deflected at different angles, resulting in the dispersion of different wavelength components along the optical axis and their convergence at different positions along the axis. Subsequently, the light carrying wavelength and spatial information enters an optical field imaging component composed of an optical field acquisition lens and a microlens array. The optical field acquisition lens primarily focuses the light. The microlens array separates light rays propagating in different directions. Each microlens unit receives light from a specific area of ​​the optical field acquisition lens and guides light rays from different directions to different pixel areas of the imaging detector. In this way, the imaging detector records not only the spatial position information of the light rays in a single exposure but also the angular distribution of the light rays through the correspondence between pixels and microlenses, thus simultaneously obtaining both the spatial position and angular distribution of the light rays—i.e., aliased spectral light field information.

[0087] The panchromatic imaging channel and the diffraction optics channel have parallel optical paths, each imaging the same target scene, meaning that both channels observe the exact same target scene. The panchromatic imaging channel includes a front lens group, which is used to acquire a high spatial resolution panchromatic image of the target. The resulting panchromatic image is only registered and fused pixel-level with the spectral cube of the diffraction-optical field channel in the digital domain, re-imaged, and transmitted to the target surface of the second imaging detector. The panchromatic imaging channel does not include dispersive elements or microlens arrays. This design ensures that light propagating in the panchromatic channel does not undergo spectral dispersion or have its angular information separated. This allows the second imaging detector to directly receive the integrated intensity information of the target scene; that is, each pixel records the total intensity of all wavelengths of light at the corresponding object point, without distinguishing spectral composition or light direction. This results in a panchromatic image with a higher spatial resolution than the diffraction optics channel, because the detector pixels in the panchromatic channel are entirely used to record spatial information, while the detector pixels in the diffraction optics channel need to simultaneously record both spatial and angular information, resulting in a relatively low spatial sampling rate.

[0088] Please see Figure 1 The present embodiment provides a hyperspectral imaging system that integrates diffraction and light field. Its core lies in the dual-path parallel snapshot acquisition of the diffraction optical channel and the panchromatic imaging channel, combined with advanced computational imaging algorithms, to replace the existing mechanical scanning and achieve high-stability and high-efficiency hyperspectral imaging.

[0089] The system's optical components consist of a diffraction optical channel and a panchromatic imaging channel. The diffraction optical channel's optical path sequentially includes a diffraction primary mirror, a light field acquisition lens, a microlens array, and a first imaging detector. The diffraction primary mirror, as the system's front-end optical element, employs a diffraction optical design, its core characteristic being axial dispersion. When incident light from the target scene passes through the diffraction primary mirror, light components of different wavelengths disperse along the optical axis and converge at different positions on the optical axis, thus achieving initial spectral separation. Subsequently, light carrying wavelength and spatial information enters the light field imaging component, composed of the light field acquisition lens and the microlens array. The light field acquisition lens converges the light, while the microlens array, located near its focal plane, separates light from different propagation directions and guides it to different pixel areas of the first imaging detector. This structure allows the system to simultaneously record the spatial position and angular distribution of light in a single exposure, obtaining aliased spectral light field information including four-dimensional light field information. The panchromatic imaging channel's optical path is set in parallel with the diffraction optical channel, imaging and unifying the target scene. The panchromatic imaging channel comprises a front lens group and a second imaging detector. The front lens group is used to clearly image the target scene onto the second imaging detector. Crucially, the panchromatic imaging channel does not contain any dispersive elements or microlens arrays; therefore, the second imaging detector directly receives the integrated intensity information of all wavelengths of light from the target scene, thus obtaining a high spatial resolution panchromatic image with a spatial resolution far exceeding that of the diffractive optical channel. This dual-channel parallel design ensures the synchronicity and snapshot characteristics of data acquisition, fundamentally avoiding mechanical scanning.

[0090] The system's processing section is implemented by a processor, responsible for jointly calculating and reconstructing the data acquired from the dual channels. The processor first receives aliased spectral light field information from the first imaging detector and a panchromatic image from the second imaging detector. Depth estimation is the first crucial step. The processor decodes the aliased spectral light field information, specifically by analyzing the dot matrix image formed by the microlens array on the first imaging detector. Based on the mapping relationship between each microlens and its subordinate pixels, the direction of light is deduced, thereby extracting multiple sets of sub-aperture images from different viewpoints. Subsequently, the pixel displacement of the same spatial feature point in each sub-aperture image between adjacent images, i.e., disparity, is calculated to obtain an initial disparity map. This disparity map is then converted into an initial depth map based on the camera's geometric model. To improve the accuracy of depth estimation in weakly textured regions, this invention introduces a panchromatic image as a spatial structure prior. The processor performs high-precision registration between the panchromatic image and the central viewpoint image extracted from the aliased spectral light field information, establishing a pixel-level correspondence. Then, constrained by the rich texture details contained in the registered panchromatic image, the initial depth map is corrected by an optimization algorithm to fill in the missing information areas and smooth the noise, and finally output a high-precision and spatially consistent scene depth map.

[0091] The next step is digital refocusing. Based on the obtained scene depth map and the dispersive characteristics of the diffractive primary mirror (i.e., the correspondence between wavelength and the position of the sharp imaging plane), the processor determines a series of refocusing plane positions corresponding to the center wavelengths within the target spectral range. For each refocusing plane position, the processor performs coordinate transformation on the four-dimensional light field data according to the light field propagation model, simulating the process of light reconverging on that plane. Then, the transformed light field data is integrated to synthesize the two-dimensional projected image corresponding to that position, i.e., the blurred spectral slice. Due to diffraction dispersion, on the refocusing plane corresponding to a specific wavelength, only the light signal of that wavelength is relatively clear, while the light signals of other wavelengths are in a defocused and blurred state. Therefore, each blurred spectral slice is the result of multi-band signal superposition.

[0092] Finally, the spectral decoupling step aims to recover a sharp hyperspectral data cube from the aliased and blurred spectral slices. The processor constructs a linear model of the system imaging, representing the blurred spectral slices as the result of convolving the hyperspectral data cube to be solved with a depth-varying three-dimensional point spread function. To solve this ill-conditioned inverse problem, the processor establishes an inversion objective function that includes data fidelity and regularization terms. Crucially, this invention incorporates panchromatic image information as a strong spatial constraint into the regularization term. Specifically, the intensity information of the panchromatic image constrains the sum of radiance of the reconstructed hyperspectral data cube across all spectral bands to be consistent with that of the panchromatic image. The processor employs an iterative optimization algorithm to solve this objective function, simultaneously decoupling sharp spatial images across all spectral bands from a set of blurred spectral slices under the guidance of the spatial constraints of the panchromatic image, ultimately outputting a high-quality hyperspectral data cube.

[0093] In summary, this implementation method successfully achieves snapshot-style hyperspectral imaging without mechanical scanning by integrating diffraction optics and light field imaging with a computational reconstruction algorithm guided by panchromatic prior information, significantly improving the stability and timeliness of the system in demanding application scenarios.

[0094] The present invention relates to a hyperspectral imaging system and method that integrates diffraction and light field fusion, which aims to solve the problems of low system stability and slow imaging speed caused by mechanical scanning components in the prior art. It is particularly suitable for vibration-sensitive applications and applications requiring high timeliness, such as space remote sensing and on-orbit scientific payloads, such as capturing spectral information of cloud movement or ocean surface changes in Earth observation.

[0095] The system hardware employs a dual-channel parallel snapshot structure. In the diffraction optics channel, a diffraction primary mirror, a light field acquisition lens, a microlens array, and a first imaging detector are sequentially arranged along the optical path. The diffraction primary mirror, acting as the front optical element, operates based on the principle of diffraction and possesses axial dispersion characteristics, causing incident light of different wavelengths to disperse and converge at different positions along the optical axis. The light then enters the light field imaging component, where it is converged by the light field acquisition lens. The microlens array separates the light rays from different propagation directions and guides them to different pixel regions of the first imaging detector, thus simultaneously recording the spatial position and angular distribution of the light in a single exposure, obtaining aliased spectral light field information. The panchromatic imaging channel operates independently and in parallel with the diffraction optics channel, featuring a front lens group and a second imaging detector. The panchromatic imaging channel does not contain dispersive elements or a microlens array; the second imaging detector directly receives the integrated intensity information of the target scene, obtaining a high spatial resolution panchromatic image. This dual-channel design ensures the synchronization of data acquisition and snapshot characteristics, eliminating the need for mechanical scanning.

[0096] Data processing is performed by the processor, which receives data acquired from both channels. In the depth estimation stage, the processor first decodes the aliased spectral light field information, analyzes the dot matrix image formed by the microlens array on the first imaging detector, and extracts multiple sets of sub-aperture images from different perspectives based on the mapping relationship between microlenses and pixels. It calculates the pixel displacement of the same spatial feature point between adjacent images in the sub-aperture images, generates an initial disparity map, and converts it into an initial depth map based on a geometric optics model. To improve depth estimation accuracy, the processor performs high-precision registration between the panchromatic image and the center-view image extracted from the aliased spectral light field information, establishing a pixel-level correspondence. Using the texture details of the panchromatic image as a constraint, it uses an optimization algorithm to correct the initial depth map, filling in the missing depth information in weak texture or occluded areas, smoothing noise, and outputting a high-precision scene depth map.

[0097] In the digital refocusing stage, the processor maps the target spectral band range to the optical axis direction based on the scene depth map and the dispersive characteristics of the diffraction primary mirror, determining the refocusing plane positions corresponding to each center wavelength. For each refocusing plane position, a coordinate transformation is performed on the four-dimensional light field data based on the light field propagation model to simulate the process of light reconverging on that plane. The transformed light field data is then integrated to synthesize a two-dimensional projected image, i.e., a blurred spectral slice. Due to diffraction dispersion, only the light signal corresponding to the wavelength in each blurred spectral slice is in quasi-focus, while other wavelengths are out of focus and blurred, forming a multi-band aliasing.

[0098] In the spectral decoupling stage, the processor constructs a linear model of the system imaging, representing the blurred spectral slice as the convolution result of the hyperspectral data cube to be reconstructed and a depth-varying three-dimensional point spread function. An inversion objective function is established, including data fidelity and regularization terms. Information from the panchromatic image is incorporated into the regularization term as a spatial constraint, and the intensity information of the panchromatic image is used to constrain the sum of radiance of each band reconstructed from the hyperspectral data cube. An iterative optimization algorithm is used to solve the inversion objective function. Under the spatial constraints of the panchromatic image, the spatial images of all spectral bands are simultaneously decoupled from the blurred spectral slice, outputting the hyperspectral data cube.

[0099] The implementation process of this invention achieves snapshot-type hyperspectral imaging without mechanical scanning through hardware fusion and computational reconstruction, thereby improving the stability and real-time performance of the system in harsh scenarios.

[0100] In the depth estimation process, the aliased spectral light field information is first converted into usable optical data through decoding. The decoding operation is based on the mapping relationship between the microlens array and the pixels of the imaging detector. By analyzing the light direction information of the pixel cluster corresponding to each microlens, multiple sets of sub-aperture images with small differences in viewing angle are reconstructed. These sub-aperture images are essentially fragments of the same scene captured from different angles. For example, in space remote sensing applications, they can present the stereoscopic viewing angle differences of cloud structures.

[0101] When calculating the pixel displacement of the same spatial feature point in a sub-aperture image set, a feature matching algorithm is used to identify corresponding points in adjacent images, and an initial disparity map is generated by measuring the pixel coordinate differences. Since the disparity value is inversely proportional to the scene depth, the disparity map is converted into an initial depth map based on the baseline distance and focal length parameters in the camera's geometric model. In marine monitoring scenarios, this process can initially reflect the three-dimensional undulation characteristics of the ocean wave surface.

[0102] To improve depth estimation accuracy, the panchromatic image is pixel-level registered with the center-view image extracted from the aliased spectral light field information. The registration process employs feature point matching and deformation model correction to align the high-frequency texture details of the panchromatic image with the light field data space. Using the texture of the registered panchromatic image as a constraint, the initial depth map is corrected using a Markov random field optimization algorithm. This algorithm leverages the structural priors provided by the panchromatic image in weakly textured regions (such as flat sea surfaces) to fill in areas lacking depth information, while simultaneously suppressing noise and outputting a scene depth map with high spatial consistency.

[0103] In the digital refocusing stage, the refocusing plane position is determined based on the scene depth map and the dispersion characteristics of the diffraction primary mirror. The dispersion equation of the diffraction primary mirror maps the target spectral band to the optical axis, calculating the sharp imaging plane corresponding to each center wavelength. For example, in vegetation monitoring, the near-infrared band convergence plane is axially offset from the visible light band. For each refocusing plane position, coordinate transformation is performed on the four-dimensional light field data based on the light field propagation model to simulate the light path propagation behavior of light reconverging on the virtual plane.

[0104] By integrating the transformed light field data at the refocusing plane, a two-dimensional projected image, i.e., a blurred spectral slice, is synthesized. The integration process is equivalent to accumulating the light intensities from all directions arriving at the plane. Due to diffraction dispersion, specific wavelengths are in quasi-focus on the corresponding refocusing plane, while other wavelengths form defocused blur spots, resulting in each blurred spectral slice containing multi-band aliasing signals. This characteristic allows for the simultaneous capture of characteristic spectral aliasing information from different minerals in mineral identification scenarios.

[0105] In the spectral decoupling stage, a linear model of the system imaging is constructed, representing the blurred spectral slices as the convolution result of a hyperspectral data cube and a three-dimensional point spread function. The point spread function varies with depth, characterizing the blurring properties of the optical system in different bands. When establishing the inversion objective function, the data fidelity term requires the reconstructed result to match the observed data, and the regularization term introduces spatial constraints. The intensity information of the panchromatic image is incorporated into the regularization term as a key constraint; its physical meaning is that the pixel value of the panchromatic image is equal to the sum of the radiant intensities of each spectral band.

[0106] Iterative optimization algorithms, such as the conjugate gradient method or variable splitting algorithm, are used to solve the inversion objective function. During the optimization process, the spatial gradient information of the panchromatic image guides the direction of spectral decoupling, suppressing aliasing while preserving edge details. Finally, spatial images of all spectral bands are simultaneously decoupled, forming a hyperspectral data cube. In environmental monitoring applications, this cube can simultaneously provide the spatial distribution and spectral characteristics of water pollutants.

[0107] Data acquisition in the diffraction optics channel begins with the modulation of incident light by the diffraction master mirror. The master mirror, through micro-nano structures, generates phase delay, causing differentiated refraction of light at different wavelengths, resulting in spectral dispersion distributed along the optical axis. The light carrying dispersion information is converged by the light field acquisition lens and then incident on a microlens array. Each microlens unit separates the received light according to its propagation direction and projects it onto different pixel areas of the first imaging detector, achieving simultaneous recording of spatial and angular information. This design allows for the simultaneous capture of the spectral reflectance characteristics of a part's surface in industrial inspection.

[0108] The panchromatic imaging channel employs an optical path structure parallel to the diffractive optical channel, with a front lens group imaging the target scene onto a second imaging detector. Since there are no dispersive elements or microlens arrays within the channel, the detector pixels directly receive the integrated intensity of all wavelengths, resulting in a high spatial resolution panchromatic image. In agricultural remote sensing scenarios, this image can provide fine structural information about crop canopies, offering a spatial reference for spectral reconstruction. The synchronized exposure mechanism of the two channels eliminates temporal errors, making it suitable for observing rapidly changing natural phenomena.

[0109] This invention provides a specific implementation of a hyperspectral imaging system that fuses diffraction and light field. Considering space remote sensing applications, the system employs a dual-channel parallel snapshot structure. The diffraction optical channel sequentially includes a diffraction primary mirror, a light field acquisition lens, a microlens array, and a first imaging detector. The diffraction primary mirror, based on diffraction optics, modulates the phase of light waves through micro-nano structures, causing different wavelengths of light to disperse and converge at different positions along the optical axis. After convergence by the light field acquisition lens, the microlens array separates light rays propagating in different directions to corresponding pixel areas of the first imaging detector. A single exposure can record four-dimensional light field data containing spatial and angular information, forming aliased spectral light field information. The panchromatic imaging channel runs parallel to the diffraction optical channel, equipped with a front lens group and a second imaging detector. This channel contains no dispersive elements or microlens arrays. The second imaging detector directly receives the integrated intensity information of the target scene and outputs a high spatial resolution panchromatic image. This dual-channel design allows for simultaneous data acquisition on a space remote sensing platform, avoiding vibration errors introduced by mechanical scanning, and is suitable for real-time spectral monitoring of dynamic scenes such as cloud movement in Earth observation.

[0110] During the data processing stage, the processor receives dual-channel data. For depth estimation, the processor first analyzes the dot matrix image formed by the microlens array on the first imaging detector, extracting a set of multi-view sub-aperture images through the mapping relationship between microlenses and pixels. It calculates the pixel displacement of the same feature point between sub-aperture images to generate an initial disparity map, which is then converted into an initial depth map based on camera geometric parameters. To improve the depth accuracy of weakly textured regions (such as flat surfaces in remote sensing images), the processor performs pixel-level registration between the panchromatic image and the center-view image extracted from the light field data. Using the texture details of the panchromatic image as a constraint, it corrects the initial depth map through an optimization algorithm, filling in missing information in occluded areas and outputting a highly consistent scene depth map. In the digital refocusing stage, the processor maps the target spectral bands to the optical axis according to the dispersion equation of the diffraction primary mirror, determining the refocusing plane position corresponding to each center wavelength. Based on the light field propagation model, it performs coordinate transformation on the four-dimensional light field data, simulating the convergence process of light rays at each refocusing plane, and synthesizes blurred spectral slices through integration. Due to the dispersion effect, each slice is only in focus for its corresponding wavelength component, while the remaining bands are out of focus and blurred, resulting in multi-band aliasing. In the spectral decoupling stage, the processor constructs a linear imaging model of the system, representing the blurred spectral slices as the convolution result of a hyperspectral data cube and a three-dimensional point spread function, and establishing an inversion objective function containing data fidelity and regularization terms. The intensity information of the panchromatic image is introduced as a spatial constraint into the regularization term, restricting the sum of radiance of the reconstructed bands to be consistent with the panchromatic image. An iterative optimization algorithm is used to solve the objective function, simultaneously decoupling the spatial images of all bands from the blurred slices under the constraints of the panchromatic image, and finally reconstructing the hyperspectral data cube. This embodiment achieves efficient spectral imaging without mechanical scanning in remote sensing scenarios through hardware collaboration and computational reconstruction.

[0111] Another embodiment optimizes the data processing flow for high-speed dynamic scenarios such as ocean surface change monitoring. In the diffraction optics channel, the axial dispersion characteristics of the diffraction primary mirror cause short-wavelength light (such as the blue light band) to converge at the near end and long-wavelength light (such as the red light band) to converge at the far end. The light field imaging component records the angular distribution of light rays through a microlens array, ensuring that a single exposure captures complete aliased spectral light field information. The front lens group of the panchromatic imaging channel adopts a large-aperture design to improve light reception efficiency, enabling the second imaging detector to acquire high signal-to-noise ratio panchromatic images even under low-light conditions. During the depth estimation stage, the processor employs a Markov random field-based optimization algorithm to address the weak texture characteristics of ocean scenes. It uses the details of sea surface wave texture in the panchromatic image as a priori to correct the errors in light field depth estimation in uniform sea surface areas. During digital refocusing, the processor automatically adjusts the refocusing plane sequence based on the scene depth map. For example, for changes in seawater depth, it prioritizes processing bands that characterize surface features. In spectral decoupling, the inversion objective function incorporates spatial gradient constraints from the panchromatic image, enhancing the ability of spectral decoupling to preserve edge information and avoiding blurring of details such as ocean fronts. This embodiment, through adaptive algorithm optimization, achieves high spatiotemporal resolution spectral data reconstruction in ocean observations, effectively capturing the spectral characteristics of transient phenomena such as algal bloom migration or oil film diffusion.

Claims

1. A hyperspectral imaging system that fuses diffraction and light field, characterized in that, include: The aliased spectral light field information of the target scene after dispersion by the diffraction primary mirror is acquired through the diffraction optical channel. The aliased spectral light field information includes the spatial and angular information of the light rays. At the same time, a high spatial resolution panchromatic image of the target scene is acquired through an independent panchromatic imaging channel. Using the spatial structure prior provided by the panchromatic image, depth estimation is performed on the aliased spectral light field information to obtain a scene depth map; Based on the scene depth map and diffraction dispersion characteristics, a series of refocusing plane positions corresponding to different center wavelengths are determined; The aliased spectral light field information is digitally refocused at the refocusing plane position to generate a set of blurred spectral slices distributed along the optical axis. Using the panchromatic image as a spatial constraint, a joint deconvolution operation is performed on the blurred spectral slices to decouple and reconstruct the hyperspectral data cube of the target scene.

2. The hyperspectral imaging system for diffraction and light field fusion according to claim 1, characterized in that, The step of using the spatial structure prior provided by the panchromatic image to perform depth estimation on the aliased spectral light field information to obtain a scene depth map includes: The aliased spectral light field information is decoded to extract a set of sub-aperture images from multiple different perspectives; Calculate the pixel displacement of the same spatial feature point in the sub-aperture image set between adjacent sub-aperture images to obtain an initial disparity map; Convert the initial disparity map into an initial depth map; The panchromatic image is registered with the central viewpoint image extracted from the aliased spectral light field information with high precision to establish a pixel-level correspondence. Using the texture details of the registered panchromatic image as constraints, the initial depth map is corrected through an optimization algorithm to fill in the missing depth information caused by weak textures or occluded areas, smooth depth noise, and output a high-precision and spatially consistent scene depth map.

3. The hyperspectral imaging system for diffraction and light field fusion according to claim 2, characterized in that, The step of digitally refocusing the aliased spectral light field information at the refocusing plane position to generate a set of blurred spectral slices distributed along the optical axis includes: Based on the dispersion equation of the diffraction primary mirror, the preset spectral band range is mapped to the optical axis direction, and the position of the refocusing plane corresponding to each spectral channel is calculated. For each of the refocusing plane positions, the coordinate transformation of the four-dimensional light field data is performed according to the light field propagation model to simulate the process of light rays reconverging at the refocusing plane position; The transformed light field data is integrated at the refocusing plane position to synthesize the two-dimensional projection image corresponding to that position, i.e., the blurred spectral slice; In this process, on the refocusing plane corresponding to a specific wavelength, the light signal of that wavelength is in a quasi-focused state, while the light signals of other wavelengths are blurred due to defocusing, so that each of the blurred spectral slices is an aliasing of multi-band signals.

4. The hyperspectral imaging system for diffraction and light field fusion according to claim 3, characterized in that, The step of using the panchromatic image as a spatial constraint to perform joint deconvolution operations on the blurred spectral slices, decoupling and reconstructing the hyperspectral data cube of the target scene, includes: A linear model of system imaging is constructed, in which the blurred spectral slice is expressed as the convolution result of the hyperspectral data cube to be reconstructed and the three-dimensional point spread function that varies with depth; Establish an inversion objective function that includes a data fidelity term and a regularization term; The information of the panchromatic image is introduced into the regularization term as a spatial constraint, wherein the intensity information of the panchromatic image is used to constrain the sum of radiance of the reconstruction results of the hyperspectral data cube in each spectral band; An iterative optimization algorithm is used to solve the inversion objective function. Under the spatial constraints of the panchromatic image, the spatial images of all spectral bands are simultaneously decoupled from the blurred spectral slices, and the hyperspectral data cube is finally output.

5. The hyperspectral imaging system for diffraction and light field fusion according to claim 4, characterized in that, The acquisition of the aliased spectral light field information of the target scene after dispersion by the diffraction primary mirror through the diffraction optical channel includes: The incident light of the target scene passes through the diffraction mirror. According to its axial dispersion characteristics, the diffraction mirror disperses light components of different wavelengths along the optical axis and converges them at different positions on the optical axis. Subsequently, the light carrying wavelength and spatial information enters the light field imaging component, which consists of a light field acquisition lens and a microlens array; The microlens array separates light rays from different propagation directions and guides them to different pixel regions of the imaging detector, thereby simultaneously recording the spatial position and angular distribution of light rays in a single exposure and obtaining the aliased spectral light field information.

6. The hyperspectral imaging system for diffraction and light field fusion according to claim 5, characterized in that, The high spatial resolution panchromatic image of the target scene is acquired through an independent panchromatic imaging channel, wherein: The optical path of the panchromatic imaging channel is set in parallel with the optical path of the diffractive optical channel, and they respectively image the same target scene. The panchromatic imaging channel includes a front lens group for imaging the target scene onto a second imaging detector; The panchromatic imaging channel does not include dispersive elements and microlens arrays, allowing the second imaging detector to directly receive the integrated intensity information of the target scene, thereby obtaining the panchromatic image with a spatial resolution higher than that of the diffraction optical channel.

7. A hyperspectral imaging system for fusing diffraction and light field, used to perform the hyperspectral imaging system for fusing diffraction and light field as described in any one of claims 1 to 6, characterized in that, The system includes: The diffraction optical channel has a diffraction primary mirror, a light field acquisition lens, a microlens array and a first imaging detector arranged sequentially in its optical path to acquire aliased spectral light field information. A panchromatic imaging channel has a front lens group and a second imaging detector arranged in its optical path to acquire high spatial resolution panchromatic images. The optical path of the panchromatic imaging channel is parallel to the optical path of the diffractive optical channel. The processor is configured to: receive the aliased spectral light field information and the panchromatic image; use the panchromatic image to perform depth estimation on the aliased spectral light field information to obtain a scene depth map; perform digital refocusing based on the scene depth map to generate a blurred spectral slice; and use the panchromatic image as a spatial constraint to perform joint deconvolution operation on the blurred spectral slice to reconstruct a hyperspectral data cube.

Images