Infrared snapshot type spectral imaging system and spectral reconstruction method

Abstract

The invention relates to a spectrum imaging system and a spectrum reconstruction method, in particular to an infrared snapshot type spectrum imaging system and a spectrum reconstruction method, and solves the technical problems that an existing infrared spectrum imaging system is complex in structure and large in volume mass. According to the infrared snapshot type spectral imaging system provided by the invention, the on-chip pixel-level infrared broadband spectral modulation array and the infrared area array detector are integrated, and the infrared broadband spectral modulation array performs spectral modulation on an incident beam, so that the infrared area array detector can obtain hyperspectral data of a to-be-detected target through single exposure; a push-broom imaging mode on which a traditional dispersion element depends is broken through, the imaging speed is high, the structure is simple, and the size and the mass are small. According to the spectrum reconstruction method provided by the invention, the spectrum transfer matrix is constructed through the transmittance curve of the spectrum channel array, and meanwhile, the dictionary learning, the generalized cross validation algorithm and the Tikhonov regularization algorithm are combined, so that complex operation is avoided, and efficient and high-precision reconstruction of the spectrum can be realized.

Description

Technical Field

[0001] This invention relates to spectral imaging systems and spectral reconstruction methods, specifically to an infrared snapshot-type spectral imaging system and spectral reconstruction method. Background Technology

[0002] Spectral imaging technology is a technique that uses multiple channels to detect and image the spectrum of a target object. It decomposes the light signal of the target object into different wavelengths through optical systems or algorithms, forming a three-dimensional data cube containing spatial and spectral data. In recent years, spectral imaging technology has shown a trend towards miniaturization, integration, and high-speed real-time processing. With the continuous advancement of spectral technology, spectral imaging technology plays an important role in more and more cutting-edge fields, constantly promoting the development of scientific research and industrial applications.

[0003] The infrared band covers an area of ​​approximately 0.7μm-14μm. Infrared spectral imaging technology is widely used in many fields due to its excellent characteristics such as good concealment, strong anti-interference ability, high-precision identification, and all-weather operation. Traditional infrared spectral imaging systems are mainly based on dispersive or Fourier interferometric spectroscopy, which requires pushbroom to obtain a complete dataset. Their system structure is complex and bulky, making it difficult to meet the development requirements of high real-time performance, lightweight design, and integration.

[0004] Snapshot-type spectral imaging technology has significant advantages in terms of acquisition speed and high integration design, and has developed rapidly in recent years. However, current research on snapshot-type spectral imaging technology mainly focuses on the visible light band, while research on the infrared band is relatively scarce, which limits its promotion and application in infrared spectral imaging technology. Summary of the Invention

[0005] The purpose of this invention is to solve the technical problems of existing infrared spectral imaging systems having complex structures and large volumes and masses, and to provide an infrared snapshot-type spectral imaging system and a spectral reconstruction method.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] An infrared snapshot-type spectral imaging system is characterized by comprising a front objective lens, an infrared broadband spectral modulation array, and an infrared array detector arranged sequentially along the transmission optical path of the incident beam, as well as a spectral reconstruction module.

[0008] The front objective lens is used to focus the incident light beam generated by the target under test onto the infrared broadband spectral modulation array.

[0009] The infrared broadband spectral modulation array includes a periodically arranged array of K spectral channels. The infrared broadband spectral modulation array is used to spectrally modulate the incident beam to obtain K modulation signals. Each spectral channel array includes N×N spectral channels arranged in an array and having different spectral responses, where N is an integer and N≥2.

[0010] The infrared array detector includes K pixel arrays corresponding to K spectral channel arrays respectively. The infrared array detector is used to detect and image K modulation signals to obtain K spectral signals. The size of each spectral channel in the spectral channel array is the same as or an integer multiple of the pixel size in the pixel array. Each spectral channel corresponds to a pixel block in the pixel array.

[0011] Where K=(a×b) / (N×N) / L, a and b are the number of pixels in the length and width directions of the infrared array detector (4) respectively, and L is the number of pixels in a pixel block;

[0012] The spectral channels of the K spectral channel array in the infrared broadband spectral modulation array are aligned and integrated one-to-one with the pixel blocks of the K pixel array in the infrared array detector.

[0013] The input of the spectral reconstruction module is connected to the output of the infrared array detector, and is used to perform spectral reconstruction on K spectral signals respectively to obtain a three-dimensional spectral image.

[0014] Furthermore, the spectral channel includes an infrared anti-reflection layer, a top reflective layer, an intermediate dielectric layer, and a bottom reflective layer stacked sequentially, wherein the infrared anti-reflection layer is used to receive the incident light beam;

[0015] The intermediate medium layers of the N×N spectral channels have different thicknesses to achieve different spectral responses through the Fabry-Perot interference effect.

[0016] Furthermore, the infrared antireflection layer is a composite antireflection film layer.

[0017] Furthermore, the infrared array detector is a cooled short-wave infrared detector, a cooled mid-wave infrared detector, a cooled long-wave infrared detector, or an uncooled long-wave detector.

[0018] This invention also provides a spectral reconstruction method using the aforementioned infrared snapshot spectral imaging system, characterized by the following steps:

[0019] Step 1: Calibrate the infrared snapshot spectral imaging system to obtain the transmittance curves of the N×N spectral channels of the K-channel array in the infrared broadband spectral modulation array.

[0020] Step 2: Based on the transmittance curves of the N×N spectral channels of the K spectral channel array in the infrared broadband spectral modulation array, calculate the corresponding spectral transfer coefficients respectively, and construct the spectral transfer matrices of the K spectral channel arrays accordingly.

[0021] Step 3: The incident beam generated by the target is focused by the front objective lens and transmitted to the infrared broadband spectral modulation array. The infrared broadband spectral modulation array modulates the beam to obtain K modulation signals and transmits them to the infrared array detector.

[0022] Step 4: The infrared array detector detects and images the K modulation signals to obtain K spectral signals and sends them to the spectral reconstruction module;

[0023] Step 5: The spectral reconstruction module, based on dictionary learning and regularization, performs spectral reconstruction on the K spectral signals according to the spectral transfer matrix of the K spectral channel array, and obtains K reconstructed spectral curves, thereby obtaining a three-dimensional spectral image.

[0024] Furthermore, step 5 specifically includes:

[0025] Step 5.1: The spectral reconstruction module calculates the gray-level response matrix of each of the K spectral signals based on the spectral transfer matrix of the K spectral channel array.

[0026] Step 5.2: In the public dataset, select the infrared spectroscopy dataset, and then use the dictionary learning method to construct a dictionary matrix based on the infrared spectroscopy dataset;

[0027] Step 5.3: Based on the dictionary matrix, discretize the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals to obtain the discretized modulation matrix; then use the Tikhonv regularization algorithm to solve for the sparse vector based on the discretized modulation matrix and the gray-level response matrix of the spectral signal.

[0028] Step 5.4: Based on the dictionary matrix and sparse vectors, perform spectral reconstruction using the following formula to obtain the reconstructed spectral curve of the spectral signal:

[0029]

[0030] in, To reconstruct the spectral curve, It is a dictionary matrix. It is a sparse vector;

[0031] Step 5.5: Repeat steps 5.3-5.4 until all K spectral signals are traversed to obtain the reconstructed spectral curves of the K spectral signals. Then, stitch them together to obtain a three-dimensional spectral image.

[0032] Furthermore, step 5.3 specifically includes:

[0033] Step 5.3.1: Based on the dictionary matrix, discretize the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals using the following formula to obtain the discretized modulation matrix:

[0034]

[0035] in, For the discretized modulation matrix, Let be the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals;

[0036] Step 5.3.2: Perform singular value decomposition on the discretized modulation matrix to obtain the singular values ​​of its diagonal matrix, and then define the filter factor based on the singular values:

[0037]

[0038] in, , These are the filter factor and singular value of the k-th spectral channel in the spectral channel array corresponding to the spectral signal, respectively. k is an integer, and 1≤k≤n, where n is the number of spectral channels in the spectral channel array, n=N×N; For regularization parameters;

[0039] Step 5.3.3: Using the generalized cross-validation algorithm, the regularization parameter is calculated based on the discretized modulation matrix, the gray-level response matrix of the spectral signal, and the filter factor using the following formula. :

[0040]

[0041] in, This represents the generalized cross-validation algorithm. The optimal solution obtained by solving for the regularization parameters. This is the grayscale response matrix;

[0042] Step 5.3.4: Using the Tikhonv regularization algorithm, based on the discretized modulation matrix, the gray-level response matrix of the spectral signal, and the regularization parameter, the sparse vector is solved using the following formula. :

[0043]

[0044] in, Describes the optimization function. This represents the L2 norm.

[0045] Further, in step 5.2, the dictionary matrix is ​​obtained by training using a dictionary learning method with the following formula:

[0046]

[0047] in, This is an infrared spectral dataset. Let m be the m-th spectrum in the infrared spectral dataset, and p be the number of spectra in the infrared spectral dataset. , is a sparse representation matrix. For using dictionary matrix express sparse vectors; Sparsity; Indicates satisfaction, Describing the Frobenius norm, This represents the L0 norm.

[0048] Further, in step 5.1, the gray-level response matrices of the K calculated spectral signals are calculated using the following formulas:

[0049]

[0050] in, Gray-scale response matrix The i-th element in the array represents the gray value of the i-th pixel block in the pixel array corresponding to the spectral signal, where i is an integer and 1≤i≤N×N; , These represent the minimum and maximum values ​​of the spectral bands corresponding to the spectral channel array of the spectral signal, respectively. The wavelength of the incident beam is Spectral signal at time; Spectral transfer matrix The i-th element in the equation represents the wavelength of the incident beam. The spectral transfer coefficient of the i-th spectral channel in the spectral channel array corresponding to the spectral signal at time; This represents the noise of the i-th pixel block in the pixel array corresponding to the spectral signal; For the infrared array detector (4), at the incident beam wavelength of Quantum efficiency at that time.

[0051] Furthermore, step 2 specifically involves:

[0052] Step 2.1: Calculate the corresponding spectral transfer coefficients based on the transmittance curves of the N×N spectral channels in the K-channel infrared broadband spectral modulation array.

[0053] Step 2.2: Define the center pixel of each of the K pixel arrays. Using the center pixel as the center, sort the N×N pixel blocks of the K pixel arrays in the infrared array detector to obtain K reconstructed spectral image periods.

[0054] Step 2.3: Sort the spectral transfer coefficients of the N×N spectral channels of the K-channel array according to the order of the corresponding pixel blocks in the reconstructed spectral image period to obtain the spectral transfer matrix of the K-channel array.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. The present invention provides an infrared snapshot type spectral imaging system, which uses an infrared broadband spectral modulation array to spectrally modulate the incident beam, so that the infrared array detector can acquire the hyperspectral data of the target in a single exposure. This breaks through the traditional push-broom imaging mode that relies on dispersive elements, and has fast imaging speed, simple structure, and small size and mass.

[0057] 2. The present invention provides an infrared snapshot type spectral imaging system that integrates an on-chip pixel-level infrared broadband spectral modulation array and an infrared external array detector. With the help of a spectral reconstruction module, it is suitable for short-wave, mid-wave and long-wave infrared bands, and can dynamically identify infrared stealth targets and efficiently acquire the infrared spectral information of the target to be measured.

[0058] 3. The spectral reconstruction method provided by this invention obtains the transmittance curve of the spectral channel array in the infrared broadband spectral modulation array through calibration, and then constructs a spectral transfer matrix based on the transmittance curve, thereby improving the accuracy of the spectral transfer matrix and thus improving the accuracy of subsequent spectral reconstruction. At the same time, a dictionary matrix for spectral reconstruction is constructed based on the dictionary learning method, and the Tikhonov regularization parameter is calculated using the generalized cross-validation (GCV) algorithm, avoiding complex calculations and achieving efficient spectral reconstruction. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of the spectral channel array in an embodiment of the present invention;

[0061] Figure 3 This is a cross-sectional view of the spectral channel in an embodiment of the present invention;

[0062] Figure 4 This is a flowchart of a method according to an embodiment of the present invention;

[0063] Figure 5 This is a schematic diagram of the transmittance curves of the nine spectral channels obtained in step 1 of an embodiment of the present invention.

[0064] Figure 6 This is a comparison chart of the reconstructed spectral curve obtained in step 5.4 of this embodiment and the reconstructed spectral curve obtained by the existing generalized cross-validation algorithm.

[0065] The annotations in the attached figures are explained as follows:

[0066] 1-Target under test, 2-Front objective lens, 3-Infrared broadband spectral modulation array, 4-Infrared array detector, 5-Spectral reconstruction module, 6-Infrared anti-reflection layer, 7-Top reflective layer, 8-Intermediate medium layer, 9-Bottom reflective layer. Detailed Implementation

[0067] The infrared snapshot-type spectral imaging system and spectral reconstruction method proposed in this invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are merely used to explain the technical principles of this invention and are not intended to limit the scope of protection of this invention.

[0068] An infrared snapshot-type spectral imaging system, such as Figure 1 As shown, the system includes a front objective lens 2, an infrared broadband spectral modulation array 3, and an infrared external array detector 4, arranged sequentially along the optical path of the incident beam, as well as a spectral reconstruction module 5. The front objective lens 2 is used to focus the incident beam generated by the target 1 onto the infrared broadband spectral modulation array 3. The infrared broadband spectral modulation array 3 includes a periodically arranged array of K spectral channels. The infrared broadband spectral modulation array 3 is used to spectrally modulate the incident beam to obtain K modulation signals, such as... Figure 2 As shown, each spectral channel array includes N×N spectral channels arranged in an array and having different spectral responses. In this embodiment, N=3. The infrared array detector 4 includes K pixel arrays corresponding to the K spectral channel arrays. The infrared array detector 4 is used to detect and image the K modulation signals to obtain K spectral signals. The size of each spectral channel in the spectral channel array is the same as or an integer multiple of the pixel size in the pixel array. Each spectral channel corresponds to a pixel block in the pixel array. Here, K is an integer, and K=a×b / (N×N) / L, where a and b are the number of pixels in the length and width directions of the infrared array detector 4, respectively, and L is the number of pixels in a pixel block.

[0069] The spectral channels of the K spectral channels in the infrared broadband spectral modulation array 3 are aligned and integrated with the pixel blocks of the K pixel array in the infrared external array detector 4. The input of the spectral reconstruction module 5 is connected to the output of the infrared external array detector 4, and is used to perform spectral reconstruction on the K spectral signals respectively to obtain a three-dimensional spectral image.

[0070] like Figure 3As shown, the spectral channels include an infrared antireflection layer 6, a top reflective layer 7, an intermediate dielectric layer 8, and a bottom reflective layer 9 stacked sequentially. The infrared antireflection layer 6 is used to receive the incident light beam. The intermediate dielectric layers 8 of the 3×3 spectral channels have different thicknesses to achieve different spectral responses through the Fabry-Perot interference effect. In this embodiment, the infrared antireflection layer 6 adopts a composite antireflection film layer to improve spectral transmittance.

[0071] The infrared broadband spectral modulation array 3 can be directly fabricated on the photosensitive surface of the infrared array detector 4 using semiconductor processing technology, or it can be first fabricated on an infrared silicon transparent substrate and then aligned and bonded to the pixel array in the infrared array detector 4 via physical bonding or mechanical structure. The infrared array detector 4 can be a cooled short-wave infrared detector, a cooled mid-wave infrared detector, a cooled long-wave infrared detector, or an uncooled long-wave infrared detector. In this embodiment, the infrared array detector 4 is a cooled mid-wave infrared detector.

[0072] This embodiment provides an infrared snapshot-type spectral imaging system that operates in the mid-infrared band. The incident beam from the target 1 is transmitted through the front objective lens 2 to the infrared broadband spectral modulation array 3 for spectral modulation. Then, it is detected by the infrared external array detector 4, and finally, spectral reconstruction is performed by the spectral reconstruction module 5 to obtain a three-dimensional spectral image. The spectral reconstruction module 5 uses a dictionary learning method to train the dictionary matrix for spectral reconstruction, and then uses the dictionary matrix to discretize the spectral transfer matrix. After the infrared external array detector 4 acquires the spectral signal, the three-dimensional spectral image is obtained through a Tikhonov regularization algorithm. Specifically, the Tikhonov regularization algorithm uses a generalized cross-validation algorithm conforming to dictionary learning to calculate its regularization parameters.

[0073] This embodiment also provides a spectral reconstruction method, employing the aforementioned infrared snapshot spectral imaging system, such as... Figure 4 As shown, it includes the following steps:

[0074] Step 1: Calibrate the above-mentioned infrared snapshot-type spectral imaging system to obtain the transmittance curves of the 3×3 spectral channels of the K-channel array in the infrared broadband spectral modulation array 3, as shown below. Figure 5 The figure shows the transmittance curves of 3×3=9 spectral channels of a spectral channel array.

[0075] Step 2: Based on the transmittance curves of the 3×3 spectral channels of the K spectral channel array in the infrared broadband spectral modulation array 3, calculate the corresponding spectral transfer coefficients, and construct the spectral transfer matrices of the K spectral channel arrays accordingly. Specifically:

[0076] Step 2.1: Calculate the corresponding spectral transfer coefficients based on the transmittance curves of the 3×3 spectral channels of the K spectral channels in the infrared broadband spectral modulation array 3.

[0077] Step 2.2: Define the center pixel of each of the K pixel arrays. Using the center pixel as the center, sort the 3×3 pixel blocks of the K pixel arrays in the infrared array detector 4 to obtain K reconstructed spectral image periods.

[0078] Step 2.3: Sort the spectral transfer coefficients of the 3×3 spectral channels of the K-channel array according to the order of the corresponding pixel blocks in the reconstructed spectral image period to obtain the spectral transfer matrix of the K-channel array.

[0079] Step 3: The incident beam generated by the target 1 is converged by the front objective lens 2 and transmitted to the infrared broadband spectral modulation array 3. The infrared broadband spectral modulation array 3 modulates the beam to obtain K modulation signals and transmits them to the infrared array detector 4.

[0080] Step 4: The infrared array detector 4 detects and images the K modulation signals, obtains K spectral signals, and sends them to the spectral reconstruction module 5.

[0081] Step 5: The spectral reconstruction module 5, based on dictionary learning and regularization, performs spectral reconstruction on the K spectral signals according to the spectral transfer matrix of the K spectral channel array, obtaining K reconstructed spectral curves, thus producing a three-dimensional spectral image. Specifically:

[0082] Step 5.1: Spectral reconstruction module 5 calculates the gray-level response matrix of each of the K spectral signals based on the spectral transfer matrix of the K spectral channel array using the following formula:

[0083]

[0084] in, Gray-scale response matrix The i-th element in the array represents the gray value of the i-th pixel block in the pixel array corresponding to the spectral signal, where i is an integer and 1≤i≤N×N; , These represent the minimum and maximum values ​​of the spectral bands corresponding to the spectral channel array of the spectral signal, respectively. The wavelength of the incident beam is Spectral signal at time; Spectral transfer matrix The i-th element in the equation represents the wavelength of the incident beam. The spectral transfer coefficient of the i-th spectral channel in the spectral channel array corresponding to the spectral signal at time; This represents the noise of the i-th pixel block in the pixel array corresponding to the spectral signal; For the infrared array detector (4), at the incident beam wavelength of Quantum efficiency at that time.

[0085] The above formula can be simplified to matrix form:

[0086]

[0087] in, It is a spectral signal. This is the noise matrix of the pixel array corresponding to the spectral signal. Noise matrix The i-th element in.

[0088] Step 5.2: In the public dataset, select the infrared spectroscopy dataset, and then use the dictionary learning method to train the dictionary matrix based on the infrared spectroscopy dataset using the following formula:

[0089]

[0090] in, This is an infrared spectral dataset. Let p be the m-th spectrum in the infrared spectrum dataset, p be the number of spectra in the infrared spectrum dataset, and n be the number of spectral channels in the spectral channel array, n = 3 × 3;

[0091] Let be a dictionary matrix, where , representing a dictionary matrix The first in There are elements, where l is the number of dictionaries in the dictionary matrix;

[0092] , is a sparse representation matrix. For using dictionary matrix express sparse vectors;

[0093] The sparsity is represented by a smaller value, indicating greater sparsity and stronger structural constraints.

[0094] Indicates satisfaction, This represents the Frobenius norm (or F-norm for short). Describing the L0 norm, , , , These are the matrix spaces over the real number field.

[0095] In step 5.2, a dictionary matrix is ​​constructed through dictionary learning to select dictionary elements from the infrared spectral dataset that meet the requirements of flexibility and adaptability, thereby obtaining a suitable dictionary matrix to approximate the spectral information. The infrared spectral dataset is an existing publicly available dataset.

[0096] Step 5.3: Based on the dictionary matrix, discretize the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals to obtain the discretized modulation matrix; then, using the Tikhonv regularization algorithm, solve for the sparse vector based on the discretized modulation matrix and the gray-level response matrix of the spectral signal. Specifically:

[0097] Step 5.3.1: Based on the dictionary matrix, discretize the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals using the following formula to obtain the discretized modulation matrix:

[0098]

[0099] in, For the discretized modulation matrix, Let be the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals;

[0100] Step 5.3.2: Perform singular value decomposition on the discretized modulation matrix to obtain the singular values ​​of its diagonal matrix, and then define the filter factor based on the singular values:

[0101]

[0102] in, , These are the filter factor and singular value of the k-th spectral channel in the spectral channel array corresponding to the spectral signal, respectively, where k is an integer and 1≤k≤n; For regularization parameters;

[0103] Step 5.3.3: Using the generalized cross-validation algorithm, the regularization parameter is calculated based on the discretized modulation matrix, the gray-level response matrix of the spectral signal, and the filter factor using the following formula. :

[0104]

[0105] in, This represents the generalized cross-validation algorithm. The optimal solution obtained by solving for the regularization parameters;

[0106] Step 5.3.4: Using the Tikhonv regularization algorithm, based on the discretized modulation matrix, the gray-level response matrix of the spectral signal, and the regularization parameter, solve for the sparse vector using the following formula. :

[0107]

[0108] in, Describes the optimization function. This represents the L2 norm.

[0109] In step 5.3, the Tikhonv regularization algorithm is used to transform the linear relationship of the spectral reconstruction matrix into an optimization equation for solving sparse vectors. Simultaneously, since the regularization parameter controls the sparsity and smoothness of the solution, a generalized cross-validation algorithm is employed to calculate the result when... The regularization parameter corresponding to the minimum value is the optimal regularization parameter.

[0110] Step 5.4: Based on the dictionary matrix and sparse vectors, perform spectral reconstruction using the following formula to obtain the reconstructed spectral curve of the spectral signal:

[0111]

[0112] in, To reconstruct the spectral curve;

[0113] Step 5.5: Repeat steps 5.3-5.4 until all K spectral signals are traversed to obtain the reconstructed spectral curves of the K spectral signals. Then, stitch them together to obtain a three-dimensional spectral image.

[0114] In step 5, a method combining dictionary learning and sparse approximation is used to achieve spectral reconstruction. First, a dictionary matrix is ​​constructed, and the spectral information is approximated by the dictionary matrix and sparse vectors. Then, the Tikhonv regularization algorithm is used to solve for the sparse vectors, transforming the linear relationship of the matrix into an optimization equation to solve for the sparse vectors, thus completing efficient and high-precision spectral reconstruction, which meets the requirements of lightweight and real-time performance of infrared spectral imaging systems.

[0115] like Figure 6 As shown, the reconstructed spectral curves obtained by existing generalized cross-validation algorithms exhibit distortion when dealing with bimodal problems, and the center wavelength also has a certain shift. This embodiment is significantly superior to the generalized cross-validation algorithm. When the PSNR (Peak Signal-to-Noise Ratio) is greater than 30dB, both the human visual system and conventional detection instruments can directly perform analysis. The reconstructed spectral curves obtained in this embodiment have a PSNR greater than 31dB and an RMSE (Root Mean Square Error) less than 0.24, exhibiting low reconstruction error and good reconstruction accuracy.

[0116] This embodiment provides an infrared snapshot-type spectral imaging system and spectral reconstruction method. It utilizes on-chip spectral modulation technology and constructs an infrared broadband spectral modulation array 3 by combining pixel-level multi-channel broadband spectral modulation and narrowband spectral reconstruction principles. Then, a dictionary matrix for spectral reconstruction is constructed using a dictionary learning method. Finally, based on the Tikhonov regularization optimization algorithm, the regularization parameters are calculated using a generalized cross-validation algorithm to complete the spectral reconstruction, achieving efficient and high-precision spectral reconstruction.

Claims

1. An infrared snapshot-type spectral imaging system, characterized in that: It includes a front objective lens (2), an infrared broadband spectral modulation array (3), and an infrared array detector (4) arranged sequentially along the incident beam transmission optical path, as well as a spectral reconstruction module (5). The front objective lens (2) is used to focus the incident beam generated by the target (1) to the infrared broadband spectral modulation array (3). The infrared broadband spectral modulation array (3) includes a periodically arranged array of K spectral channels. The infrared broadband spectral modulation array (3) is used to spectrally modulate the incident beam to obtain K modulation signals. Each spectral channel array includes N×N spectral channels arranged in an array and having different spectral responses, where N is an integer and N≥2. The infrared array detector (4) includes K pixel arrays corresponding to K spectral channel arrays respectively. The infrared array detector (4) is used to detect and image K modulation signals to obtain K spectral signals. The size of each spectral channel in the spectral channel array is the same as or an integer multiple of the pixel size in the pixel array. Each spectral channel corresponds to a pixel block in the pixel array. Where K=(a×b) / (N×N) / L, a and b are the number of pixels in the length and width directions of the infrared array detector (4) respectively, and L is the number of pixels in a pixel block; The spectral channels of the K spectral channel array in the infrared broadband spectral modulation array (3) are aligned and integrated one by one with the pixel blocks of the K pixel array in the infrared array detector (4). The input of the spectral reconstruction module (5) is connected to the output of the infrared array detector (4) and is used to perform spectral reconstruction on K spectral signals respectively to obtain a three-dimensional spectral image.

2. The infrared snapshot spectral imaging system according to claim 1, characterized in that: The spectral channel includes an infrared anti-reflection layer (6), a top reflective layer (7), an intermediate dielectric layer (8), and a bottom reflective layer (9) stacked in sequence. The infrared anti-reflection layer (6) is used to receive the incident light beam. The intermediate medium layer (8) of the N×N spectral channels has different thicknesses to achieve different spectral responses through the Fabry-Perot interference effect.

3. The infrared snapshot spectral imaging system according to claim 2, characterized in that: The infrared anti-reflection layer (6) is a composite anti-reflection film layer.

4. An infrared snapshot-type spectral imaging system according to any one of claims 1-3, characterized in that: The infrared array detector (4) is a cooled short-wave infrared detector, a cooled mid-wave infrared detector, a cooled long-wave infrared detector, or an uncooled long-wave detector.

5. A spectral reconstruction method, employing an infrared snapshot-type spectral imaging system as described in any one of claims 1-4, characterized in that, Includes the following steps: Step 1: Calibrate the infrared snapshot type spectral imaging system to obtain the transmittance curves of the N×N spectral channels of the K spectral channel array in the infrared broadband spectral modulation array (3); Step 2: Based on the transmittance curves of the N×N spectral channels of the K spectral channel array in the infrared broadband spectral modulation array (3), calculate the corresponding spectral transfer coefficients respectively, and construct the spectral transfer matrix of the K spectral channel arrays accordingly. Step 3: The incident beam generated by the target (1) is focused by the front objective lens (2) and transmitted to the infrared broadband spectral modulation array (3). The infrared broadband spectral modulation array (3) modulates the beam to obtain K modulation signals and transmits them to the infrared array detector (4). Step 4: The infrared array detector (4) detects and images the K modulation signals to obtain K spectral signals and sends them to the spectral reconstruction module (5). Step 5, Spectral Reconstruction Module (5) Based on dictionary learning and regularization, the K spectral signals are reconstructed according to the spectral transfer matrix of the K spectral channel array, resulting in K reconstructed spectral curves, thereby obtaining a three-dimensional spectral image.

6. The spectral reconstruction method according to claim 5, characterized in that, Step 5 specifically involves: Step 5.1, Spectral Reconstruction Module (5): Calculate the grayscale response matrix of each of the K spectral signals based on the spectral transfer matrix of the K spectral channel array. Step 5.2: In the public dataset, select the infrared spectroscopy dataset, and then use the dictionary learning method to construct a dictionary matrix based on the infrared spectroscopy dataset; Step 5.3: Based on the dictionary matrix, discretize the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals to obtain the discretized modulation matrix; then use the Tikhonv regularization algorithm to solve for the sparse vector based on the discretized modulation matrix and the gray-level response matrix of the spectral signal. Step 5.4: Based on the dictionary matrix and sparse vectors, perform spectral reconstruction using the following formula to obtain the reconstructed spectral curve of the spectral signal: ； in, To reconstruct the spectral curve, It is a dictionary matrix. It is a sparse vector; Step 5.5: Repeat steps 5.3-5.4 until all K spectral signals are traversed to obtain the reconstructed spectral curves of the K spectral signals. Then, stitch them together to obtain a three-dimensional spectral image.

7. The spectral reconstruction method according to claim 6, characterized in that, Step 5.3 specifically involves: Step 5.3.1: Based on the dictionary matrix, discretize the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals using the following formula to obtain the discretized modulation matrix: ； in, For the discretized modulation matrix, Let be the spectral transfer matrix of the spectral channel array corresponding to one of the spectral signals; Step 5.3.2: Perform singular value decomposition on the discretized modulation matrix to obtain the singular values ​​of its diagonal matrix, and then define the filter factor based on the singular values: ； in, , These are the filter factor and singular value of the k-th spectral channel in the spectral channel array corresponding to the spectral signal, respectively. k is an integer, and 1≤k≤n, where n is the number of spectral channels in the spectral channel array, n=N×N; For regularization parameters; Step 5.3.3: Using the generalized cross-validation algorithm, based on the discretized modulation matrix, the gray-level response matrix of the spectral signal, and the filter factor, calculate the regularization parameter using the following formula. : ； in, This represents the generalized cross-validation algorithm. The optimal solution obtained by solving for the regularization parameters. This is the grayscale response matrix; Step 5.3.4: Using the Tikhonv regularization algorithm, based on the discretized modulation matrix, the gray-level response matrix of the spectral signal, and the regularization parameter, the sparse vector is solved using the following formula. : ； in, Describes the optimization function. This represents the L2 norm.

8. The spectral reconstruction method according to claim 7, characterized in that, In step 5.2, the dictionary matrix is ​​obtained by training using the dictionary learning method with the following formula: ； in, This is an infrared spectral dataset. Let m be the m-th spectrum in the infrared spectral dataset, and p be the number of spectra in the infrared spectral dataset. , is a sparse representation matrix. For using dictionary matrix express sparse vectors; Sparsity; Indicates satisfaction, Describing the Frobenius norm, This represents the L0 norm.

9. The spectral reconstruction method according to claim 8, characterized in that, In step 5.1, the gray-level response matrices of the K calculated spectral signals are calculated using the following formulas: ； in, Gray-scale response matrix The i-th element in the array represents the gray value of the i-th pixel block in the pixel array corresponding to the spectral signal, where i is an integer and 1≤i≤N×N; , These represent the minimum and maximum values ​​of the spectral bands corresponding to the spectral channel array of the spectral signal, respectively. The wavelength of the incident beam is Spectral signal at time; Spectral transfer matrix The i-th element in the equation represents the wavelength of the incident beam. The spectral transfer coefficient of the i-th spectral channel in the spectral channel array corresponding to the spectral signal at time; This represents the noise of the i-th pixel block in the pixel array corresponding to the spectral signal; For the infrared array detector (4), at the incident beam wavelength of Quantum efficiency at that time.

10. A spectral reconstruction method according to any one of claims 5-9, characterized in that, Step 2 is as follows: Step 2.1: Calculate the corresponding spectral transfer coefficients based on the transmittance curves of the N×N spectral channels of the K spectral channel array in the infrared broadband spectral modulation array (3). Step 2.2: Define the center pixel of each of the K pixel arrays. Using the center pixel as the center, sort the N×N pixel blocks of the K pixel array in the infrared array detector (4) to obtain K recombined spectral image periods. Step 2.3: Sort the spectral transfer coefficients of the N×N spectral channels of the K-channel array according to the order of the corresponding pixel blocks in the reconstructed spectral image period to obtain the spectral transfer matrix of the K-channel array.

Images