Wide-field spectral imaging self-calibration method and system based on differentiable physics-driven

Abstract

This invention relates to the field of spectral imaging technology, and more particularly to a wide-field-of-view spectral imaging self-calibration method and system based on differentiable physics. The self-calibration method first constructs a spectral imaging device consisting of a filter array and an image sensor; it then combines a differentiable physics forward computation model with the actual light transmission physical process inside the filter to obtain scalar spectral measurements; it acquires raw transmission spectrum data under sparse angle incidence; it constructs a joint optimization objective function to iteratively solve for the real microstructural parameters, accurately retrieving the filter's thin film physical thickness and refractive index coefficient; it dynamically corrects the spectral sensing measurement matrix under arbitrary large-angle incidence conditions in a wide field of view, completing spectral imaging decoding and reconstruction. This invention can accurately compensate for the spectral blue shift effect caused by large-angle tilted incidence, effectively weaken the spectral distortion problem caused by changes in the incident angle and micro-processing deviations, and eliminate pseudo-redshift defects and structural imaging artifacts during spectral reconstruction.

Description

Technical Field

[0001] This invention belongs to the field of spectral imaging technology, and particularly relates to a wide-field spectral imaging self-calibration method and system based on differentiable physics. Background Technology

[0002] Spectral imaging technology can detect the radiation intensity of targets across continuous frequency bands and acquire a three-dimensional data cube containing two-dimensional spatial texture and one-dimensional continuous spectral characteristics. It is widely used in many engineering fields such as biomedical diagnostics, precision agriculture, and autonomous driving. Currently, portable smart terminals place stringent requirements on the size, weight, and power consumption of spectral imaging optical systems. Miniaturization, low power consumption, and integration have become important development directions for spectral imaging devices. Traditional spatial dispersive and narrowband interferometric spectral imaging systems have complex beam-splitting structures and are highly sensitive to mechanical vibration, making them unsuitable for high-dynamic, portable detection scenarios. Therefore, snapshot-based computational spectral imaging technology based on a Broadband Multispectral Filter Array (BMSFA) is currently the mainstream technology for achieving chip-level miniaturization of spectral imaging systems. This technology employs a front-end optical encoding combined with back-end computational reconstruction architecture, integrating the beam-splitting structure onto the surface of the image sensor. It relies on an observation matrix to complete spectral signal inversion, and the accuracy of the observation matrix directly determines the quality of the spectral reconstruction.

[0003] In industrial fabrication and wide-field imaging applications, BMSFA exhibits significant drawbacks. Firstly, manufacturing errors exist in the device fabrication process, resulting in substantial deviations between actual physical parameters and ideal design parameters. Secondly, during wide-field operation, the oblique incidence of edge rays generates a cosine modulation effect, causing a blue shift in the transmission spectrum. This leads to a severe mismatch between the actual measurement matrix and the theoretical design matrix at the wide-field edges. Using a static calibration matrix for spectral decoding introduces mapping errors, resulting in distortions such as false redshift and structural artifacts in the reconstructed spectrum, significantly reducing imaging accuracy.

[0004] To address the spectral distortion problem caused by model mismatch, existing calibration techniques mainly fall into two categories. The first is the precision mechanical exhaustive scanning calibration method. This method utilizes a multi-axis displacement stage and goniometer to perform point-by-point full-band spectral response tests on sensor pixels at different incident angles. While this method can obtain accurate spectral response data, the number of calibration parameters increases dramatically with device resolution, resulting in time-consuming and costly calibration processes, making it unsuitable for mass production of chips. The second category is data-driven calibration methods, including deep learning algorithms and linear numerical interpolation methods. Deep learning algorithms rely on sparse measured data to train the network, lacking physical constraints and exhibiting poor generalization ability for incident angle data outside the training set, easily generating non-physical oscillation artifacts. Linear interpolation methods cannot fit the nonlinear variation of optical parameters, resulting in low calibration accuracy and poor physical fidelity. Summary of the Invention

[0005] In view of this, the present invention aims to provide a wide field-of-view spectral imaging self-calibration method and system based on differentiable physics, which eliminates spectral distortion and imaging artifacts caused by large-angle tilted incidence by constructing a differentiable physics forward calculation model and combining it with the actual optical transmission physical process.

[0006] To achieve the above objectives, the technical solution created by this invention is implemented as follows: A wide-field-of-view spectral self-calibration method includes the following steps: S1: Construct a spectral imaging device, including a filter array and an image sensor. The filter array is coupled and integrated into the photosensitive surface of the image sensor. Each filter is composed of multiple layers of optical thin films stacked one after another. All filters have uniform material dispersion properties. S2: Based on the multilayer film transport matrix, a differentiable physical forward calculation model is constructed. The transmittance of the transverse electric mode and the transverse magnetic mode is independently solved by the differentiable physical forward calculation model to obtain scalar spectral measurement values. S3: Using at least one filter and the corresponding image sensor pixels as superpixel calibration samples, collect the raw data of the measured transmission spectrum corresponding to the incident light at a sparse angle; S4: Based on the globally shared refractive index coefficient and independent thin film physical thickness of all filters, and combining scalar spectral measurements with measured transmission spectrum data, a joint optimization objective function is constructed. The gradient optimization method is used for iterative solution to obtain the true microstructure parameters of all filters. S5: Using real microstructure parameters, calculate the transmittance curves at different incident angles across the entire field of view. Input the mask formed by all transmittance curves into the spectral reconstruction algorithm to obtain the corrected spectral sensing measurement matrix. Based on the corrected spectral sensing measurement matrix, complete the spectral imaging decoding and reconstruction to eliminate spectral distortion and imaging artifacts caused by large-angle tilted incidence.

[0007] Furthermore, step S2 includes the following steps: S21: Calculate the first using Cauchy's dispersion equation j Thin film at different incident wavelengths refractive index coefficient : ; in, , , For the first j The Cauchy coefficients of the thin film to be inverted; S22: Calculate the... j The phase thickness of the thin film corresponds to the transverse electric mode and the transverse magnetic mode. for: ; in, The imaginary unit, For the first j The extinction coefficient of the thin film, Let be the complex refractive index of the thin film. For the first j The physical thickness of the thin film. For the first j Cosine value of the incident angle of the layer light rays; S23: Phase thickness Substitute the multilayer film transmission matrix and calculate the transmittance of the transverse electric mode and the transverse magnetic mode independently; S24: The transmittance of the transverse electric mode and the transverse magnetic mode is incoherently superimposed to obtain scalar spectral measurement values.

[0008] Furthermore, in step S3, the sparse angles include at least the normal incident angle of the vertically incident filter and a set of tilted incident angles.

[0009] Furthermore, in step S4, the objective function is jointly optimized. for: ; in, The total number of superpixel calibration samples. s To calibrate the sample number, For the first s Each calibration sample at wavelength The measured transmittance was as follows. Transmittance calculated by the differentiable physical forward computation model. The Cauchy coefficient to be inverted, For the first s A set of physical thickness parameters corresponding to each thin film layer of a calibration sample; within These are all parameters to be optimized.

[0010] Furthermore, in step S4, the actual microstructure parameters include the physical thickness and refractive index coefficient of each thin film layer of the filter.

[0011] A wide-field-of-view spectral self-calibration system for implementing the aforementioned self-calibration method includes: A spectral imaging device includes a filter array and an image sensor, wherein the filter array is coupled and integrated into the photosensitive surface of the image sensor; The scalar spectral measurement acquisition module is used to construct a differentiable physical forward calculation model. The transmittance of the transverse electric mode and transverse magnetic mode spectra is independently calculated through the differentiable physical forward calculation model to obtain scalar spectral measurement values. The spectral acquisition module is used to acquire the raw data of the measured transmission spectrum when light is incident on the superpixel calibration sample; wherein, the superpixel calibration sample includes at least one filter and the corresponding pixels of the image sensor; The joint optimization module is used to construct a joint optimization objective function based on the globally shared refractive index coefficient and independent film physical thickness of all filters, combined with scalar spectral measurements and measured transmission spectral raw data. The gradient optimization method is used to iteratively solve the parameters to obtain the true microstructure parameters of all filters. The spectral imaging reconstruction module is used to calculate the transmittance curves at different incident angles using real microstructure parameters. All transmittance curves are used to form a mask and input into the spectral reconstruction algorithm to obtain the corrected spectral sensing measurement matrix. The spectral imaging decoding and reconstruction is completed based on the corrected spectral sensing measurement matrix.

[0012] Furthermore, the filter array employs broadband multispectral filters, and the image sensor is a complementary metal-oxide-semiconductor image sensor. Compared with the prior art, the present invention can achieve the following beneficial effects: This invention combines a differentiable physical forward computation model with the actual optical transmission physics within the device, replacing the traditional mechanical exhaustive scanning calibration that relies on high-density continuous angle sampling. It uses measured sparse angle transmission spectral data to iteratively solve for the true microstructural parameters, accurately retrieving the filter film thickness, dispersion coefficient, and other intrinsic microstructural parameters of the device, thus meeting the batch calibration needs of industrial-grade devices. Furthermore, this invention utilizes the truly retrieved microstructural parameters to dynamically correct the spectral sensing measurement matrix under wide-field-of-view, arbitrarily large-angle incident conditions. This accurately compensates for the spectral blue shift effect caused by large-angle tilted incident light, effectively mitigating spectral distortion caused by changes in incident angle and micro-processing deviations, and eliminating pseudo-redshift defects and structural imaging artifacts during spectral reconstruction. Attached Figure Description

[0013] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic flowchart of the wide-field spectral imaging self-calibration method based on differentiable physics driven by the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0015] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0016] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0017] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0018] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] like Figure 1As shown, an embodiment of the present invention provides a wide-field-of-view spectral imaging self-calibration method based on differentiable physics, comprising the following steps: S1: Construct a spectral imaging device, including a filter array and an image sensor. The filter array is coupled and integrated into the photosensitive surface of the image sensor. Each filter is composed of multiple layers of optical thin films stacked one after another. All filters have uniform material dispersion properties.

[0020] In this embodiment, the filter array is a broadband multispectral filter array, and the image sensor is a complementary metal-oxide-semiconductor (CMOS) image sensor. The broadband multispectral filter array is fabricated using a multilayer optical thin-film process. Each broadband multispectral filter is composed of multiple layers of optical thin films stacked sequentially. The broadband multispectral filter array is coupled and integrated onto the photosensitive surface of the CMOS image sensor, allowing incident light to penetrate perpendicularly through the array before reaching the photosensitive pixels of the CMOS image sensor. All broadband multispectral filters possess uniform material dispersion properties, ensuring consistent material processing properties while preserving the random thickness variations inherent in micro / nano fabrication.

[0021] This step utilizes a multilayer optical thin-film interference structure to complete optical encoding and spatial multiplexing of the incident spectrum; it achieves modulation and aliasing of spectral signals in different bands through the thin-film interference resonance effect, compressing high-dimensional spectral information into two-dimensional spatial pixels, thus completing the front-end optical encoding process. The spectral imaging device in this embodiment has a compact structure, eliminating the need for complex beam-splitting optical paths and meeting the requirements for chip-level miniaturization integration.

[0022] S2: Based on the multilayer film transport matrix, a differentiable physical forward calculation model is constructed. The transmittance of the transverse electric mode and the transverse magnetic mode is independently calculated through the differentiable physical forward calculation model to obtain scalar spectral measurement values.

[0023] A vectorized differentiable physical forward computation model is constructed based on the multilayer film transfer matrix method, transforming the physical process of thin-film optical transmission into a differentiable computation node adapted to a deep learning framework. This differentiable physical forward computation model embeds multiple optical physical constraints, specifically including: the optical path difference correction relationship corresponding to the incident angle, the thin film phase thickness calculation relationship, and the complex refractive index solution logic.

[0024] Step S2 includes the following steps: S21: Calculate the first using Cauchy's dispersion equation j Thin film at different incident wavelengths refractive index coefficient : ; in, , , For the first j The Cauchy coefficients of the thin film to be inverted; By introducing the Cauchy dispersion equation as a way to characterize dispersion parameters, the broadband dispersion characteristics of optical materials are characterized by parameterization and dimensionality reduction, which reduces the number of optimization parameters and avoids the computational redundancy caused by solving discrete wavelengths one by one.

[0025] S22: Calculate the... j The phase thickness of the thin film corresponds to the transverse electric mode and the transverse magnetic mode. for: ; in, The imaginary unit, For the first j The extinction coefficient of the thin film, Let be the complex refractive index of the thin film. For the first j The physical thickness of the thin film. For the first j Cosine value of the incident angle of the layer light rays; This formula quantifies the nonlinear coupling relationship between the incident angle, the refractive index of the medium, and the film thickness, and accurately captures the physical blue shift phenomenon caused by the shortening of the optical path difference.

[0026] S23: Phase thickness Substitute the multilayer film transmission matrix and calculate the transmittance of the transverse electric mode and the transverse magnetic mode independently.

[0027] Since the calculation of transmittance using a multilayer film transmission matrix is ​​a prior art technique, it will not be described in detail in this invention.

[0028] S24: The transmittance of the transverse electric mode and the transverse magnetic mode is incoherently superimposed to obtain scalar spectral measurement values.

[0029] For both transverse electric and transverse magnetic polarization modes, transmittance is calculated independently, and the final scalar spectral measurements are synthesized through incoherent light intensity superposition, achieving end-to-end differentiable gradient backhaul from macroscopic broadband probe spectra to microscopic physical geometric parameters. Furthermore, this differentiable physics forward computation model supports tensor operations and automatic differentiation, providing computational support for subsequent gradient optimization.

[0030] S3: Using at least one filter and the corresponding image sensor pixels as superpixel calibration samples, collect the raw data of the measured transmission spectrum corresponding to the incident light at a sparse angle.

[0031] In some embodiments, the sparse angles include at least the normal incident angle of the vertically incident filter and a set of tilted incident angles.

[0032] This embodiment uses only two incident angles: a normal incident angle at the vertical incident filter and a 30° tilted incident angle. For each calibration sample, measured transmission spectrum data at different incident angles are collected to obtain a spectral response dataset under real-world operating conditions.

[0033] This step uses a sparse angle sampling method, which significantly reduces the cost and sampling time of mechanical calibration, making it suitable for industrial mass production calibration needs.

[0034] S4: Based on the globally shared refractive index coefficient and independent film physical thickness of all filters, combined with the scalar spectral measurement value obtained in step S2 and the measured transmission spectrum raw data obtained in step S3, a joint optimization objective function is constructed, and the gradient optimization method is used to iteratively solve the problem to obtain the real microstructure parameters of all filters.

[0035] In some embodiments, the actual microstructure parameters include the physical thickness and refractive index coefficient of each thin film layer of the filter.

[0036] This embodiment relies on the process characteristics of coated devices from the same wafer and batch to decouple parameters. All filters within the batch use identical optical media with uniform material dispersion characteristics, thus all filters share the same set of refractive index coefficients. Due to the uniformity error of the coating process, the actual physical thickness of each thin film layer in different filters varies slightly, and the film thicknesses of each filter are independent and not shared. By using a parameter decomposition method of "global dispersion uniformity and local film thickness independence," the parameter degeneracy caused by the coupling of refractive index and film thickness is broken, and a unique and stable solution for the microstructure parameters is achieved through joint optimization of multiple samples.

[0037] The joint optimization objective function constructed in this embodiment for: ; in, The total number of superpixel calibration samples. s To calibrate the sample number, For the first s Each calibration sample at wavelength The measured transmittance was as follows. Transmittance calculated by the differentiable physical forward computation model. The Cauchy coefficient to be inverted, For the first s A set of physical thickness parameters corresponding to each thin film layer of a calibration sample; within These are all parameters to be optimized.

[0038] In this embodiment, the joint optimization objective function is used to quantify the residual between the measured transmission spectrum data and the scalar spectrum measurement. A gradient optimization method combined with automatic differentiation backpropagation is used to iteratively update the dispersion parameters and film thickness parameters within the joint parameters. This iteration continues until the residual converges, ultimately uniquely inverting to obtain the true microstructure parameters corresponding to each superpixel.

[0039] This embodiment employs an Adaptive Moment Estimation (Adam) optimizer, which is hardware-accelerated based on Compute Unified Device Architecture (CUDA), to perform gradient iterative optimization. This optimization scheme is existing technology and will not be elaborated upon here.

[0040] S5: Using real microstructure parameters, calculate the transmittance curves at different incident angles across the entire field of view. Input the mask formed by all transmittance curves into the spectral reconstruction algorithm to obtain the corrected spectral sensing measurement matrix. Based on the corrected spectral sensing measurement matrix, complete the spectral imaging decoding and reconstruction to eliminate spectral distortion and imaging artifacts caused by large-angle tilted incidence.

[0041] Based on the real microstructure parameters obtained from the above steps, the spectral sensing measurement matrix at any incident angle across the entire field of view is dynamically corrected, and the corrected spectral sensing measurement matrix replaces the traditional static theoretical design matrix. During wide-field imaging, the dynamic spectral sensing measurement matrix is ​​used to decode and reconstruct the spectral image, accurately compensating for the nonlinear changes in optical path difference caused by large-angle tilted incidence, correcting the spectral blue shift effect caused by tilted incidence, eliminating false redshift and structural artifacts in the reconstructed spectrum, and finally outputting high-fidelity, low-distortion three-dimensional spectral data.

[0042] This embodiment also provides a wide-field-of-view spectral imaging self-calibration system based on differentiable physics, used to implement the aforementioned self-calibration method, including: A spectral imaging device includes a filter array and an image sensor, wherein the filter array is coupled and integrated into the photosensitive surface of the image sensor; The scalar spectral measurement acquisition module is used to construct a differentiable physical forward calculation model. The transmittance of the transverse electric mode and transverse magnetic mode spectra is independently calculated through the differentiable physical forward calculation model to obtain scalar spectral measurement values. The spectral acquisition module is used to acquire the raw data of the measured transmission spectrum when light is incident on the superpixel calibration sample; wherein, the superpixel calibration sample includes at least one filter and the corresponding pixels of the image sensor; The joint optimization module is used to construct a joint optimization objective function based on the globally shared refractive index coefficient and independent thin film physical thickness of all filters, combined with scalar spectral measurements and measured transmission spectrum raw data. The gradient optimization method is used to iteratively solve the parameters to obtain the true microstructure parameters of all filters. The spectral imaging reconstruction module is used to calculate the transmittance curves at different incident angles using real microstructure parameters. All transmittance curves are used to form a mask and input into the spectral reconstruction algorithm to obtain the corrected spectral sensing measurement matrix. The spectral imaging decoding and reconstruction is completed based on the corrected spectral sensing measurement matrix.

[0043] In some embodiments, the filter array employs a broadband multispectral filter, and the image sensor is a complementary metal-oxide-semiconductor image sensor.

[0044] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0045] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A wide-field-of-view spectral imaging self-calibration method based on differentiable physics, characterized in that, Includes the following steps: S1: Construct a spectral imaging device, including a filter array and an image sensor. The filter array is coupled and integrated into the photosensitive surface of the image sensor. Each filter is composed of multiple layers of optical thin films stacked one after another. All filters have uniform material dispersion properties. S2: Based on the multilayer film transport matrix, a differentiable physical forward calculation model is constructed. The transmittance of the transverse electric mode and the transverse magnetic mode is independently solved by the differentiable physical forward calculation model to obtain scalar spectral measurement values. S3: Using at least one filter and the corresponding image sensor pixels as superpixel calibration samples, collect the raw data of the measured transmission spectrum corresponding to the incident light at a sparse angle; S4: Based on the globally shared refractive index coefficient and independent thin film physical thickness of all filters, and combining scalar spectral measurements with measured transmission spectrum data, a joint optimization objective function is constructed. The gradient optimization method is used for iterative solution to obtain the true microstructure parameters of all filters. S5: Using real microstructure parameters, calculate the transmittance curves at different incident angles across the entire field of view. Input the mask formed by all transmittance curves into the spectral reconstruction algorithm to obtain the corrected spectral sensing measurement matrix. Based on the corrected spectral sensing measurement matrix, complete the spectral imaging decoding and reconstruction to eliminate spectral distortion and imaging artifacts caused by large-angle tilted incidence.

2. The wide-field-of-view spectral imaging self-calibration method based on differentiable physics as described in claim 1, characterized in that, Step S2 includes the following steps: S21: Calculate the first using Cauchy's dispersion equation j Thin film at different incident wavelengths refractive index coefficient : ； in, , , For the first j The Cauchy coefficients of the thin film to be inverted; S22: Calculate the... j The phase thickness of the thin film corresponds to the transverse electric mode and the transverse magnetic mode. for: ； in, The imaginary unit, For the first j The extinction coefficient of the thin film, Let be the complex refractive index of the thin film. For the first j The physical thickness of the thin film. For the first j Cosine value of the incident angle of the layer light rays; S23: Phase thickness Substitute the multilayer film transmission matrix and calculate the transmittance of the transverse electric mode and the transverse magnetic mode independently; S24: The transmittance of the transverse electric mode and the transverse magnetic mode is incoherently superimposed to obtain scalar spectral measurement values.

3. The wide-field-of-view spectral imaging self-calibration method based on differentiable physics as described in claim 1, characterized in that, In step S3, the sparse angles include at least the normal incident angle of the vertically incident filter and a set of tilted incident angles.

4. The wide-field-of-view spectral imaging self-calibration method based on differentiable physics as described in claim 1, characterized in that, In step S4, the joint optimization objective function for: ； in, The total number of superpixel calibration samples. s To calibrate the sample number, For the first s Each calibration sample at wavelength The measured transmittance was as follows. Transmittance calculated by the differentiable physical forward computation model. The Cauchy coefficient to be inverted, For the first s A set of physical thickness parameters corresponding to each thin film layer of a calibration sample; within These are all parameters to be optimized.

5. The wide-field-of-view spectral imaging self-calibration method based on differentiable physics as described in claim 1, characterized in that, In step S4, the actual microstructure parameters include the physical thickness and refractive index coefficient of each thin film layer of the filter.

6. A wide-field-of-view spectral imaging self-calibration system based on differentiable physics, used to implement the self-calibration method according to any one of claims 1 to 5, characterized in that, include: A spectral imaging device includes a filter array and an image sensor, wherein the filter array is coupled and integrated into the photosensitive surface of the image sensor; The scalar spectral measurement acquisition module is used to construct a differentiable physical forward calculation model. The transmittance of the transverse electric mode and transverse magnetic mode spectra is independently calculated through the differentiable physical forward calculation model to obtain scalar spectral measurement values. The spectral acquisition module is used to acquire the raw data of the measured transmission spectrum when light is incident on the superpixel calibration sample; wherein, the superpixel calibration sample includes at least one filter and the corresponding pixels of the image sensor; The joint optimization module is used to construct a joint optimization objective function based on the globally shared refractive index coefficient and independent thin film physical thickness of all filters, combined with scalar spectral measurements and measured transmission spectrum raw data. The gradient optimization method is used to iteratively solve the parameters to obtain the true microstructure parameters of all filters. The spectral imaging reconstruction module is used to calculate the transmittance curves at different incident angles using real microstructure parameters. All transmittance curves are used to form a mask and input into the spectral reconstruction algorithm to obtain the corrected spectral sensing measurement matrix. The spectral imaging decoding and reconstruction is completed based on the corrected spectral sensing measurement matrix.

7. The wide-field-of-view spectral imaging self-calibration system based on differentiable physics as described in claim 6, characterized in that, The filter array uses broadband multispectral filters, and the image sensor is a complementary metal-oxide-semiconductor image sensor.

Images