Medium wave infrared calculation spectral imaging method based on dictionary learning and sparse reconstruction

Abstract

The invention belongs to the technical field of medium-wave infrared spectral imaging, and particularly relates to a medium-wave infrared calculation spectral imaging method based on dictionary learning and sparse reconstruction. The method comprises the following steps: S1, randomly selecting m spectral signals from a smooth spectral signal data set, and inputting the m spectral signals into a dictionary learning algorithm for dictionary learning to obtain an initial dictionary; s2, loading the gas transmittance curve and the black-body radiation spectrum signal into the initial dictionary to obtain a sparse dictionary for medium-wave infrared spectrum signal regression; s3, constructing a physical measurement matrix of a broadband optical filter group in the optical system; and S4, imaging the spectral data cube of the target scene by using an optical system to obtain imaging data, and reconstructing the spectral data cube of the target scene based on the sparse dictionary, the imaging data and the physical measurement matrix by using a sparse reconstruction algorithm. The method integrates the advantages of dictionary learning and sparse reconstruction, and has the characteristics of small data demand, short training time, high reconstruction speed and the like.

Description

Technical Field

[0001] This invention belongs to the field of mid-wave infrared spectral imaging technology, and particularly relates to a mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction. Background Technology

[0002] Compared to traditional imaging methods, a significant advantage of spectral imaging technology lies in its ability to acquire a three-dimensional data cube of a target scene—including a two-dimensional spatial image and spectral information associated with each pixel. Currently, this technology is widely used in various fields such as target detection, mineral exploration, and biomedical diagnostics. Furthermore, it provides a feasible and highly promising technical approach for monitoring greenhouse gas and pollutant emissions at the regional scale.

[0003] Despite the immense potential of spectral imaging technology across various fields, its practical application still faces numerous challenges. Traditional spectral imaging systems rely on complex optical systems and mechanical scanning, and suffer from large size and slow acquisition speeds. In contrast, computational spectral imaging technology based on spectral encoding and decoding, which has emerged in recent years, effectively alleviates the limitations of traditional spectral imaging instruments by combining the spectral encoding process of the optical system with the back-end spectral decoding process. Specifically, by using specially designed optical elements to compress and encode data cubes, multi-channel spectral data can be reconstructed from a small amount of compressed measurement data within a compressed sensing framework, providing a feasible approach to the miniaturization and weight reduction of spectral imaging instruments.

[0004] Currently, research on spectral imaging based on spectral encoding and decoding mainly focuses on the visible light band, and spectral reconstruction primarily employs deep learning methods. This data-driven spectral reconstruction algorithm achieves high accuracy within the visible light band and enables snapshot hyperspectral imaging. However, this method heavily relies on large datasets for model training. Most critically, sufficiently large datasets are lacking for most bands outside the visible light band, which limits its practical application.

[0005] In contrast, mid-wave infrared (MWIR) spectroscopy can capture rich molecular vibrational features in the 3-5 µm wavelength range, thus playing an irreplaceable role in the qualitative and quantitative analysis of gases. However, due to the unique material fingerprint information and inherent sharp absorption characteristics of infrared spectral signals, spectral reconstruction methods targeting the visible light band are difficult to apply to their reconstruction tasks. Therefore, spectral reconstruction methods targeting the visible light band often cannot achieve high-precision reconstruction of specific material fingerprint features in MWIR spectral signals. Summary of the Invention In view of this, the present invention aims to provide a mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction to solve the problems of existing technologies: 1) complex system structure, large size, and high cost; 2) need for scanning components or high-precision moving mirrors, resulting in slow response; 3) poor reconstruction accuracy of high-frequency features such as gas absorption peaks; 4) deep learning schemes rely on a large amount of training data; 5) poor system adaptability to the environment and difficulty in industrial deployment. The present invention combines dictionary learning and sparse reconstruction algorithms to realize computational spectral imaging in the MWIR band. The present invention integrates the advantages of dictionary learning and sparse reconstruction, and has the characteristics of small data requirements, short training time, and fast reconstruction speed.

[0006] To achieve the above objectives, the technical solution created by this invention is implemented as follows: A mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction specifically includes the following steps: S1: Randomly select m spectral signals from the smooth spectral signal dataset and input them into the dictionary learning algorithm to learn the dictionary and obtain the initial dictionary; S2: Load the gas transmittance curve and blackbody radiation spectrum signal into the initial dictionary to obtain a sparse dictionary for mid-wave infrared spectral signal regression. S3: Construct the physical measurement matrix of the broadband filter group in the optical system; S4: The optical system is used to image the spectral data cube of the target scene to obtain imaging data. The sparse reconstruction algorithm reconstructs the spectral data cube of the target scene based on the sparse dictionary, imaging data and physical measurement matrix.

[0007] Furthermore, in step S1, the dictionary learning algorithm includes the K-SVD dictionary learning algorithm.

[0008] Furthermore, in step S1, m is at least 50,000.

[0009] Furthermore, in step S2, the gas transmittance curves include carbon dioxide transmittance curves, sulfur dioxide transmittance curves, carbon monoxide transmittance curves, and nitrogen monoxide transmittance curves; the blackbody radiation spectrum signal is a blackbody radiation spectrum signal from 353K to 553K.

[0010] Furthermore, the optical system includes a mid-wave infrared imaging lens, a broadband filter group, and a mid-wave infrared detector arranged sequentially along the direction of light propagation.

[0011] Furthermore, a physical measurement matrix for the broadband filter group is constructed based on the spectral transmittance matrix of each broadband filter contained in the broadband filter group.

[0012] Furthermore, sparse reconstruction algorithms include orthogonal matching pursuit, basis pursuit, or minimum angle regression.

[0013] Furthermore, the smoothed spectral signal dataset includes the ARAD-1K dataset.

[0014] Compared with the prior art, the present invention can achieve the following beneficial effects: The present invention provides a mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction, which features: using only broadband filters, eliminating the need for interferometric systems or dispersive elements, resulting in a simple system structure and low cost; acquiring multi-channel information in a single image, eliminating the need for scanning, achieving fast imaging speed, and supporting snapshot imaging; the constructed sparse dictionary integrates gas absorption features and blackbody radiation features, resulting in high reconstruction accuracy and suitability for gas detection; small-sample dictionary learning avoids the reliance on large amounts of data in deep learning, requiring less data and shortening training time; the broadband filter structure is simple, compatible with existing infrared detectors, and easy to integrate and deploy in industrial applications. Attached Figure Description

[0015] 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 A schematic flowchart of the mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction described in the embodiments of the present invention; Figure 2 A schematic diagram illustrating the learning process of the initial dictionary as described in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of the optical system described in the embodiments of the present invention; Figure 4 A schematic diagram illustrating the principle of spectral signal reconstruction as described in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of the broadband filter described in an embodiment of the present invention.

[0016] Explanation of reference numerals in the attached figures: 1. Spectral data cube of the target scene; 2. Mid-wave infrared imaging lens; 3. Broadband filter group; 4. Mid-wave infrared detector; 5. Raw spectral data cube. Detailed Implementation

[0017] 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.

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0019] 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.

[0020] 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.

[0021] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] like Figure 1 As shown, the present invention provides a mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction, which specifically includes the following steps: S1: Randomly select m spectral signals from the smooth spectral signal dataset and input them into the dictionary learning algorithm to obtain an initial dictionary; S2: Load the gas transmittance curve and blackbody radiation spectral signal into the initial dictionary to obtain a sparse dictionary for mid-wave infrared spectral signal regression; S3: Construct the physical measurement matrix of the broadband filter group 3 in the optical system; S4: Use the optical system to image the spectral data cube 1 of the target scene to obtain imaging data, and the sparse reconstruction algorithm reconstructs the spectral data cube 1 of the target scene based on the sparse dictionary, imaging data and physical measurement matrix.

[0023] It should be noted that, in order to solve the sparse reconstruction problem of spectral signals, such as Figure 2As shown, this invention first utilizes a smoothed spectral signal dataset for dictionary learning to obtain an initial dictionary, i.e., a dictionary for smoothed spectral signal regression is obtained through dictionary learning. Then, a sparse dictionary suitable for mid-wave infrared spectral signal regression is constructed based on the gas's spectral transmittance curve and blackbody radiation spectrum. The sparse dictionary constructed in this way enables the smoothed spectral signal components and sharp gas absorption components to work synergistically during the sparse reconstruction process, achieving high-precision reconstruction of gas absorption spectra. The dictionary learning employs an alternating optimization strategy. In each iteration, the dictionary is first kept fixed, and the linear regression problem is solved to update the sparse encoding. Then, the sparse encoding is fixed, and the dictionary is updated to minimize the objective function.

[0024] In some embodiments, in step S1, the dictionary learning algorithm includes the K-SVD dictionary learning algorithm.

[0025] In some embodiments, in step S1, m is at least 50,000.

[0026] In some embodiments, in step S2, the gas transmittance curves include carbon dioxide transmittance curves, sulfur dioxide transmittance curves, carbon monoxide transmittance curves, and nitrogen monoxide transmittance curves; the blackbody radiation spectrum signal is a blackbody radiation spectrum signal of 353K-553K.

[0027] In some embodiments, the optical system includes a mid-wave infrared imaging lens 2, a broadband filter group 3, and a mid-wave infrared detector 4 arranged sequentially along the direction of light propagation.

[0028] It should be noted that, as Figure 3 As shown, the spectral data cube 1 of the target scene is imaged using an optical system, and its projection is obtained on the mid-wave infrared detector 4. During this process, the spectral information of the target scene is modulated and encoded using a broadband filter, thereby acquiring integrated light intensity information at the mid-wave infrared detector 4. Using the light intensity information at the corresponding pixel as the compressed measurement result, a compressed sensing algorithm is applied to obtain the reconstructed spectral data cube. The spectral signal processing flow is as follows: Figure 4 As shown, the spectrum at a certain point in the target scene is multiplied by the spectral transmittance of a broadband filter to obtain the modulated spectrum. The modulated spectra corresponding to multiple broadband filters are integrated to obtain the intensity value at the corresponding pixel. Based on the above intensity value, reconstruction is performed to obtain the reconstructed spectrum.

[0029] In some embodiments, the physical measurement matrix of the broadband filter group 3 is constructed based on the spectral transmittance matrix of each broadband filter included in the broadband filter group 3.

[0030] The broadband filter group 3 switches between the broadband filters via a filter wheel.

[0031] It should be noted that this invention employs a broadband filter based on an optical thin film for spectral encoding. This optical thin film uses an alternating structure of high and low refractive index layers, such as... Figure 5 As shown. After determining the optical thin film structure of the broadband filter, the corresponding spectral transmittance can be derived based on the characteristic matrix of the optical thin film, thereby determining the spectral response of the system, i.e., determining the physical measurement matrix. The result of multiplying the physical measurement matrix by the sparse dictionary is used as the sparse measurement matrix. The sparse reconstruction process is achieved by processing real-world data input based on the sparse measurement matrix and the orthogonal matching pursuit algorithm. Nine broadband filters are used for modulation and encoding of scene spectral information. Each broadband filter has eight optical thin films. The low-refractive-index material is magnesium fluoride, and the high-refractive-index material is titanium dioxide. The thickness of a single film layer is in the range of 100-600 nm.

[0032] Nine filters are used to modulate and encode the scene's spectral information. Each filter has eight optical thin films, with magnesium fluoride as the low-refractive-index material and titanium dioxide as the high-refractive-index material. The thickness of a single film layer ranges from 100 to 600 nm. Broadband filters can also be replaced by novel spectral modulation devices such as metasurfaces.

[0033] In some embodiments, the sparse reconstruction algorithm is an orthogonal matching pursuit algorithm, a basis pursuit algorithm, or a minimum angle regression algorithm.

[0034] In some embodiments, the smoothed spectral signal dataset includes the ARAD-1K dataset.

[0035] It should be noted that 50,000 spectral signals were randomly selected from the ARAD-1K dataset for dictionary learning, resulting in an initial dictionary for smoothing spectral signal regression. During dictionary learning, the Least Angle Regression (LARS) algorithm was used to solve the linear regression problem to achieve spectral feature selection; the dictionary was updated row-by-row. Based on the initial dictionary, blackbody radiation spectral signals from 353-553K were fused with transmittance curves of four gases (carbon dioxide, sulfur dioxide, carbon monoxide, and nitrogen monoxide) to construct a sparse dictionary for mid-wave infrared spectral signal regression. By combining this with the physical measurement matrix of broadband filter group 3, a compressed measurement matrix can be obtained. Finally, the Orthogonal Matching Pursuit (OMP) algorithm was used to reconstruct the original spectral data cube 5. Furthermore, during dictionary construction, dictionary learning algorithms such as K-Singular Value Decomposition (K-SVD) can be used instead of the LARS + row-by-row dictionary update method.

[0036] 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.

[0037] 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 mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction, characterized in that: Specifically, the steps include the following: S1: Randomly select m spectral signals from the smooth spectral signal dataset and input them into the dictionary learning algorithm to learn the dictionary and obtain the initial dictionary; S2: Load the gas transmittance curve and blackbody radiation spectrum signal into the initial dictionary to obtain a sparse dictionary for mid-wave infrared spectral signal regression. S3: Construct the physical measurement matrix of the broadband filter group in the optical system; S4: The optical system is used to image the spectral data cube of the target scene to obtain imaging data. The sparse reconstruction algorithm reconstructs the spectral data cube of the target scene based on the sparse dictionary, imaging data and physical measurement matrix.

2. The mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction according to claim 1, characterized in that: In step S1, the dictionary learning algorithm includes the K-SVD dictionary learning algorithm.

3. The mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction according to claim 1, characterized in that: In step S1, m is at least 50,000.

4. The mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction according to claim 1, characterized in that: In step S2, the gas transmittance curves include carbon dioxide transmittance curves, sulfur dioxide transmittance curves, carbon monoxide transmittance curves, and nitrogen monoxide transmittance curves; the blackbody radiation spectrum signal is a blackbody radiation spectrum signal from 353K to 553K.

5. The mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction according to claim 1, characterized in that: The optical system includes a mid-wave infrared imaging lens, a broadband filter group, and a mid-wave infrared detector arranged sequentially along the direction of light propagation.

6. The mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction according to claim 1, characterized in that: The physical measurement matrix of the broadband filter group is constructed based on the spectral transmittance matrix of each broadband filter contained in the broadband filter group.

7. The mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction according to claim 1, characterized in that: Sparse reconstruction algorithms include orthogonal matching pursuit algorithm, basis pursuit algorithm, or minimum angle regression algorithm.

8. The mid-wave infrared computational spectral imaging method based on dictionary learning and sparse reconstruction according to claim 1, characterized in that: The smoothed spectral signal dataset includes the ARAD-1K dataset.

Images