Sparse reconstruction of interferometric hyperspectral image based on adaptive threshold
Wen Jia1,2, Liu Mingwei1,2, Cui Jun1,2, Yan Shuxia1,2     
1. School of Electronics and Information Engineering, Tianjin Polytechnic University, Tianjin 300387, China;
2. Tianjin Key Laboratory of Optoelectronic Detection Technology and Systems, Tianjin 300387, China

Overview: The interference hyperspectral image data is a three-dimensional image data generated by satellite scanning by a Large Aperture Static Imaging Spectrometer (LASIS) based on the principle of push-scan Fourier transform imaging. The resolution is extremely high, and its massive amount Data poses a certain degree of difficulty for data storage and transmission over limited bandwidth channels. Therefore, it is imperative to design an efficient transmission method suitable for interfering hyperspectral data for its data characteristics. Compressed sensing, as a new theoretical framework, provides new research ideas for signal description and processing. Unlike the existing sampling theorem, the theory samples the signal using a rate much smaller than the Nyquist sampling law, and then reconstructs the original signal with high probability from these small observations. This efficient sampling method greatly reduces the sampling rate, so it has great application prospects in many research fields. Based on the traditional compressed sensing reconstruction algorithm, this paper proposes a reconstruction method for interference hyperspectral image. The interference hyperspectral image is a three-dimensional image with multi-dimensional correlation, and its interference fringes contain abundant spectral information. However, when using traditional ROMP algorithm to reconstruct the image, the absolute value of the inner product of the measurement matrix and the residual needs to be calculated. As the interference hyperspectral image has interference fringes with large fluctuations in the fixed amplitude, the variance of the calculation result of the inner product is large, which will result in too many atoms to be selected in the secondary selection according to the regularization standard in each iteration. The atomic number with higher matching degree in the subsequent phase is not selected, resulting in support. The proportion of atoms with high degree of central matching is low. This will seriously affect the reconstruction quality of interference hyperspectral, especially the interference fringe. To solve the above problems, in this paper we propose an adaptive threshold regularized orthogonal matching pursuit algorithm (ATROMP). The algorithm first uses block processing and then selects the interference fringes. Because the vertical interference fringes have strong unidirectional characteristics, the interference fringes in the images are extracted from the horizontal total variation values for adaptive sampling. Then an adaptive threshold is used in this paper to replace the quadratic selection in the ROMP algorithm. Using an adaptive threshold can not only ensure that the atomicity of each selected atom is sufficiently high, but also that multiple atoms can be properly selected each time to ensure sufficient number of cycles, to avoid the follow-up higher degree of atom missing. Compared with the traditional ROMP algorithm, a large amount of experimental datas show that the sparse reconstruction accuracy of the method can be significantly improved.

Supported by Natural Science Foundation of Tianjing (17JCQNJC01400) and National Natural Science Foundation of China (61401439, 61601323)