Hyperspectral polarization-compressed imaging and reconstruction with sparse basis optimized by particle swarm optimization

Hyperspectral polarized images are widely used in target detection, and compressive sensing has been developed to shorten the acquisition time and reduce the memory usage. However, the current systems have defects in measuring the polarization information, and current sparse bases meet problems in reconstructing process. This paper combines compressive sensing and machine learning to obtain polarized images. Both the simpler system and the sparse basis optimizing method are developed. Liquid crystal tunable filter (LCTF) serves as both hyperspectral filter and linear polarizer. The LCTF compresses the first three Stokes parameters by changing the transmission axis angle of its incident surface. The LCTF and quarter-wave plate (QWP) compress all four Stokes parameters by changing the fast axis angle of QWP. Then, the Stokes parameters can be reconstructed based on the sparse basis optimized by particle swarm optimization (PSO) algorithm. The sparse basis optimized through the spatially local images of a target in one spectral band also works for both global images of this target in many other spectral bands and images of other targets.