Hyperspectral high-resolution imaging method based on compressed sensing Fourier laminated imaging

Abstract

The invention relates to the technical field of high-resolution imaging, discloses a hyperspectral high-resolution imaging method based on compressed sensing Fourier laminated imaging, and provides a hyperspectral high-resolution rapid imaging technology based on Fourier laminated imaging. Through combination of a single-exposure hyperspectral compression imaging technology and a Fourier lamination imaging technology, hyperspectral high-resolution rapid imaging is realized. According to the method, a two-dimensional image is acquired by using a single-exposure hyperspectral compression imaging acquisition module, hyperspectral images of multiple wavebands are calculated through a deep learning algorithm, then a high-resolution image is synthesized by using a Fourier lamination imaging algorithm, and finally the high-resolution hyperspectral images of the multiple wavebands are obtained. The image reconstructed by the method provided by the invention can simultaneously meet the requirements of high resolution and high spectrum, and the method provided by the invention has the advantages of improving the imaging speed, reducing the hardware requirement standard and cost and improving the acquisition efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral high-resolution imaging, and in particular to a hyperspectral high-resolution imaging method based on compressed sensing Fourier stack imaging. Background Art

[0002] High-resolution imaging is a goal that is constantly pursued in the development of the Earth observation field. Increasing the aperture of the imaging system can improve the optical performance of the system and the resolution of the acquired image, thereby obtaining better imaging capabilities. However, as the aperture increases, the manufacturing and maintenance costs of the imaging system will also increase accordingly.

[0003] Fourier stack imaging uses a low-magnification objective to collect multiple low-resolution images at different angles, uses the recorded light intensity information as an amplitude constraint in the spatial domain, and uses the pupil function of the imaging system as a phase constraint in the frequency domain. Through alternating iterations in the spatial and frequency domains, these images are fused in the spectrum to obtain a high-resolution, large-field-of-view, and quantitative phase image.

[0004] Fourier stack imaging has high resolution and high signal-to-noise ratio, does not require interference, and is simple to implement. At the same time, it can restore relatively accurate phase information and additional aberration information, providing a new idea for long-distance, high-resolution imaging. It is of great significance to the further development of the field of high-resolution earth observation, and has important research value in military, police, and civilian fields.

[0005] However, with the development of information theory, people's demand for image acquisition continues to increase. Two-dimensional light intensity images can no longer meet all people's needs. They need to perceive the dimensional information of light wave physical quantities such as spectrum, polarization and phase. Although Fourier stack imaging has excellent performance in two-dimensional high-resolution imaging, it still faces challenges in high-spectrum imaging.

[0006] In the process of collecting hyperspectral information, it is necessary to obtain the spectral information of each band by scanning, which is a time-consuming process. For Fourier stacking imaging technology, it is necessary to move the camera to collect multiple low-resolution images from various angles and synthesize a high-resolution image, which also consumes a lot of time. Therefore, the imaging speed of Fourier stacking technology based on hyperspectral images is slow, and achieving high-speed imaging has become an important issue at present. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a high-spectral high-resolution imaging method based on compressed sensing Fourier stack imaging.

[0008] First of all, the technical solution of the present invention is as follows:

[0009] Single-exposure hyperspectral compression imaging technology only collects two-dimensional observation images and reconstructs scene spectral images through algorithms. It has fast acquisition speed, simple system structure and low cost. Among them, single-exposure hyperspectral compression imaging technology based on coded aperture can be used to solve the problem of slow hyperspectral Fourier stack imaging.

[0010] The basic principle of the single-exposure hyperspectral compression imaging system is to encode a three-dimensional spectral image and compress its spectral dimension, compress it into a two-dimensional observation image for collection, and reconstruct the original three-dimensional spectral image through this two-dimensional observation. For a three-dimensional data, the scene is first encoded by a modulation mask, where the mask consists of a series of pixels that transmit light and those that do not transmit light. The modulated three-dimensional spectral image is further dispersed by a dispersive element (such as a prism), and the images of different spectral bands are displaced in the dispersion direction and staggered with each other. Finally, they are detected by the camera sensor. The detection process is equivalent to integrating the modulated spectral image in the spectral dimension. After obtaining the compressed observation, combined with the modulation information of the mask, the detected spectral image can be reconstructed by the reconstruction algorithm.

[0011] Based on the above ideas, the steps of the hyperspectral high-resolution imaging method based on compressed sensing Fourier stack imaging proposed in the present invention are as follows:

[0012] S1, using a single-exposure hyperspectral compression imaging acquisition module to replace the camera in the traditional Fourier stack imaging system for image acquisition, wherein the single-exposure hyperspectral compression imaging acquisition module uses single-exposure hyperspectral compression imaging technology to acquire two-dimensional images;

[0013] S2, for each collected 2D image, use the deep learning algorithm to calculate the hyperspectral image of 139 bands;

[0014] S3, for each band of images, reconstruct them through Fourier stacking imaging algorithm, synthesize high-resolution images, and obtain high-resolution hyperspectral images of 139 bands;

[0015] S4. The high-resolution images of each band are synthesized into a high-spectral high-resolution image.

[0016] Preferably, when the Fourier stack imaging algorithm reconstructs the high-resolution image of each wavelength, it iterates alternately in the spatial domain and the frequency domain to obtain high-resolution complex amplitude information.

[0017] Preferably, two constraints are utilized in the high-resolution image reconstruction process for each wavelength:

[0018] In the spatial domain, a low-resolution image of one wavelength is taken as the amplitude constraint of the optimal solution;

[0019] In the frequency domain, the pupil aperture function is taken as the spectral support domain constraint of the optimal solution. This circular aperture is scanned and stacked in the spectrum to form a larger frequency passband, thereby recovering the high-frequency information. The specific iterative method can use the GS algorithm or the EPRY algorithm.

[0020] Compared with the related art, the high-spectral high-resolution imaging method based on compressed sensing Fourier stack imaging provided by the present invention has the following beneficial effects:

[0021] 1. The present invention proposes a hyperspectral high-resolution rapid imaging method based on Fourier stack imaging, which combines the Fourier stack imaging system with the single-exposure hyperspectral compression imaging technology. It not only retains the advantage of the Fourier stack imaging system in improving resolution, but also combines the function of single-exposure hyperspectral compression imaging technology in processing hyperspectral data. The reconstructed image can meet both high-resolution and hyperspectral requirements.

[0022] 2. When collecting low-resolution hyperspectral images, there is no need to collect three-dimensional hyperspectral images, only two-dimensional images are needed. Compared with the traditional hyperspectral data collection method, the hardware requirements are lowered, thereby reducing costs;

[0023] 3. For the collected two-dimensional images, low-resolution hyperspectral images of multiple bands can be quickly calculated through deep learning algorithms, which improves the efficiency of collection and saves the time required for collection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the process of high-resolution imaging for an equivalent large aperture;

[0025] Figure 2 Schematic diagram of the imaging process for each subaperture. DETAILED DESCRIPTION

[0026] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0027] Please refer to Figure 1-2 ,in, Figure 1 In the process of high-resolution imaging with equivalent large aperture, the light reflected by the object is regarded as parallel light, which is imaged on the image plane through the large aperture and then directly captured by the camera to obtain a high-resolution image.

[0028] Figure 2 For the imaging process of each sub-aperture, the light reflected by the object is parallel light. The parallel light is imaged on the image plane through the sub-aperture, and then the two-dimensional information is collected by the single-exposure hyperspectral compression imaging acquisition module. Then, the low-resolution hyperspectral image is calculated, and the aperture moves continuously along the path in the figure to realize the acquisition of multiple sets of low-resolution hyperspectral images.

[0029] The present invention utilizes multiple sub-aperture imaging to synthesize a large aperture for imaging, and the specific implementation process is as follows:

[0030] (1) Figure 2 As shown, a single sub-aperture is imaged first, and the spectral information of the single sub-aperture is modulated by the mask of the single-exposure hyperspectral compression imaging acquisition module. This information is compressed in the spectral dimension and compressed into a two-dimensional image, which is collected by the camera. Then, a low-resolution hyperspectral image of the single sub-aperture is reconstructed by a deep learning algorithm or a traditional iterative algorithm.

[0031] (2) According to Figure 2 The path in the figure is used to move the aperture position and the camera position to achieve multi-angle imaging and collect multiple sets of low-resolution high-spectral images.

[0032] (3) Extract low-resolution images of each wavelength separately, and reconstruct high-resolution images of each wavelength separately using Fourier stacking imaging algorithm. High-resolution complex amplitude information is obtained by alternating iterations in the spatial domain and the frequency domain. Two constraints are used in the reconstruction process:

[0033] In the spatial domain, a low-resolution image of one wavelength is taken as the amplitude constraint of the optimal solution;

[0034] In the frequency domain, the pupil aperture function is taken as the spectral support domain constraint of the optimal solution. This circular aperture is scanned and stacked in the spectrum to form a larger frequency passband, thereby recovering the high-frequency information. The specific iterative method can use the GS algorithm or the EPRY algorithm.

[0035] (4) The high-resolution images at each wavelength are synthesized into a high-spectral high-resolution image.

[0036] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A high-spectral high-resolution imaging method based on compressed sensing Fourier stack imaging, characterized in that: The following steps are involved: S1, using a single-exposure hyperspectral compression imaging acquisition module to replace the camera in the traditional Fourier stack imaging system for image acquisition, wherein the single-exposure hyperspectral compression imaging acquisition module uses single-exposure hyperspectral compression imaging technology to acquire two-dimensional images; S2, for each collected 2D image, use the deep learning algorithm to calculate the hyperspectral image of 139 bands; S3, for each band of images, reconstruct them through Fourier stacking imaging algorithm, synthesize high-resolution images, and obtain high-resolution hyperspectral images of 139 bands; S4. The high-resolution images of each band are synthesized into a high-spectral high-resolution image.

2. The method for high-resolution high-spectral imaging based on compressed sensing Fourier stack imaging according to claim 1, characterized in that: The single-exposure hyperspectral compression imaging acquisition module only acquires two-dimensional observation images. Its basic principle is to encode a three-dimensional spectral image and compress its spectral dimension into a two-dimensional observation image for acquisition, and reconstruct the original three-dimensional spectral image through this two-dimensional observation.

3. The method for high-resolution high-spectral imaging based on compressed sensing Fourier stack imaging according to claim 2, characterized in that: The process of collecting two-dimensional observation images by the single-exposure hyperspectral compression imaging acquisition module is: For a three-dimensional data, the scene is first encoded by a modulation mask, where the mask consists of a series of pixels that are transparent and non-transparent; The modulated three-dimensional spectral image is further dispersed by a dispersive element, and images of different spectral bands are displaced in the dispersion direction and staggered with each other. Finally, they are detected by a camera sensor. The detection process is equivalent to integrating the modulated spectral image in the spectral dimension. After obtaining the compressed observation, combined with the modulation information of the mask, the detected spectral image can be reconstructed through a reconstruction algorithm.

4. The method for high-resolution high-spectral imaging based on compressed sensing Fourier stack imaging according to claim 1, characterized in that: When reconstructing a high-resolution image of each wavelength, the Fourier stack imaging algorithm alternately iterates in the spatial domain and the frequency domain to obtain high-resolution complex amplitude information. Two constraints are used in the reconstruction process: In the spatial domain, a low-resolution image of one wavelength is taken as the amplitude constraint of the optimal solution; In the frequency domain, the pupil aperture function is used as a spectral support domain constraint for the optimal solution. This circular aperture is scanned and stacked in the spectrum to form a larger frequency passband, thereby recovering high-frequency information.

Images