The coded aperture snapshot spectral imager (CASSI) system measures multidimensional signals by compressing the original signal into a 2D snapshot, and then reconstructs the original signal by employing certain reconstruction algorithms. In recent years, the compressed sensing (CS) algorithm has been used to solve the inverse problems in the CASSI system. However, the practical application of the CS algorithm requires a balance among accuracy, robustness, speed, and stability. In this research, the CS algorithm is combined with a CASSI system with a division of focal plane polarization detector to reconstruct spatial, spectral and polarization information at the same time. The reconstruction results correspond to the reference data, and the experimental results show that the targets can be distinguished by spectral and polarization information. Moreover, in the scenes characterized by significant variations in light intensity, targets can be distinguished from the background through polarization imaging detection, indicating its advantages under certain circumstances. Inspired by genetic algorithms, a method for optimizing the coded aperture targeting certain sparse basis is also proposed. The results demonstrate that the correlation between the observation matrix and the sparse basis has decreased by 42.7%, leading to the improvement of the reconstruction effect of the spectral curves.
The signal-to-noise ratio (SNR) is an important performance evaluation index of polarization spectral imaging remote sensors. The SNR-estimation method based on the existing remote sensor is not perfect. To improve the SNR of this model, a partial detector check slant direction is presented in this study, and a polarization extinction ratio related to the internal SNR model of a typical multispectral imaging remote sensor is combined with the vector radiative transfer model to construct the atmosphere 6SV–SNR coupling model. The new result is that the central wavelength of the detection spectrum, the observation zenith angle, and the extinction ratio all affect the SNR of the remote sensor, and the SNR increases with the increase in the central wavelength of the detection spectrum. It is proved that the model can comprehensively estimate the SNR of a typical polarization multispectral imaging remote sensor under different detection conditions, and it provides an important basis for the application evaluation of such remote sensors.
31 January 2023 in IEEE Transactions on Geoscience and Remote Sensing
A polarization spectral imaging system based on dual dispersion and coded aperture can obtain multidimensional data cube containing the polarization, spectral, and spatial information of the target through snapshots, and it is a typical detection system in remote sensing. However, because the reconstruction method of the polarization spectrum images of this system is imperfect, most of the images are reconstructed through multiple exposures based on cyclic S-matrix coding, which limits its real-time performance. To resolve this problem, this study proposes a multi-slit encoding method for reconstructing multidimensional data cube with a single exposure. The method is based on the imaging characteristics of the system, and a multi-slit encoding scheme is designed so that the detector plane can obtain all of the polarization spectral information at one point of the target. A single-exposure experiment shows that the acquisition and processing speed of the system is considerably improved from 1.67 fps to 15 fps. Furthermore, the peak signal-to-noise ratio of the image reconstructed by single-exposure reconstruction reaches 34.37 dB, and the average structural similarity index is 0.912. The proposed method is proven to be effective in promoting the applicability of the spectral imaging system in real-time detection.
In the Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST) project, fiber positioners align the fiber ends to designated positions on the focal plane to capture high-quality spectra of specific celestial objects. Therefore precise measurement of fiber end positions is essential for accurate control in the closed-loop positioning system. Current measurement systems primarily rely on 2D planar methods, which cannot achieve precise positioning of fiber ends on LAMOST’s curved focal plane. However traditional 3D measurement methods often struggle to reconstruct specific targets, leading to excessive data redundancy and reduced accuracy. To address these issues, this paper proposes a novel monocular vision-based 3D measurement method for LAMOST fiber ends under both front and back illumination. In this method, coding points and a baseline ruler are placed on the focal plane; the coding points help determine camera poses, while the baseline ruler recovers the physical scale of the measurement results. Images of the coding points and fiber end-faces are first captured under front and back illumination using a single camera that takes frames with movement, and the enhanced spot-grayscale distribution model algorithm is introduced as an efficient and accurate method for spot center extraction. Subsequently, the decoding method of coding points-clustering and pattern matching method is proposed for automated recognition and decoding of the coding points. Finally, a fusion-coded structure-from-motion based 3D reconstruction method, is employed to measure the fiber end positions in 3D. Results show that the method achieves a root mean square error of 0.049 mm in fiber optic 3D measurement error. The proposed 3D measurement method for fiber ends is vital for closed-loop control in LAMOST and is also applicable to high-precision measurement of the fiber ends in other multi-object spectroscopic telescopes.