Research On Spectral Imaging Systems And Reconstruction Algorithms Based On Compressive Sensing

With the combination of imaging and spectroscopy,spectral imaging simultaneously obtains the three-dimensional data cube with two dimensions for spatial information and one dimension for spectra.The three-dimensional data cube can either be regarded as the intensity images for each spectral channel or provide the spectra for each spatial pixel.Based on the abundant information in spectral images,spectral imaging is extensively employed in fields including remote sensing,agriculture,biology,medicine,military,and etc.Different from the conventional spectral imaging with spatial or spectral scanning,compressive spectral imaging realizes non-scanning and high-efficient spectral imaging with compressive sampling and computational reconstruction based on the compressive sensing theory.In compressive spectral imaging,spectral images are first compressively captured by the optical system,with encoding in both spectral and spatial dimensions;and then spectral images are reconstructed from the sampling results.Both the sampling process and the reconstruction algorithm determine the sampling efficiency and the quality for the reconstructed spectral images.In this work,the study on the optical systems for compressive sampling and the reconstruction algorithms is conducted on the coded aperture snapshot spectral imaging(CASSI)system and the single-pixel spectral imaging system,based on the coherence theory in compressive sensing.Firstly,we propose a joint CASSI and RGB imaging system,and the corresponding pre-fusion and post-fusion reconstruction algorithms.The high fidelity in spectral dimension for CASSI and the high resolution in spatial dimension for RGB imaging are fully adopted and combined in the joint CASSI and RGB imaging system,which provides more effective information compared to only one contributed imaging system.In the pre-fusion algorithm,the system matrices for compressive sampling in two contributed imaging systems are concatenated for high-efficient reconstruction.In the post-fusion algorithm,the guided-filtering-based fusion significantly improves the spectral and spatial resolution of the reconstructed images.Secondly,we propose the dual-disperser CASSI system with RGB image sensor based on the coherence minimization principle according to the compressive sensing theory.The coded aperture and the over-complete dictionary are optimized with genetic algorithm and gradient descent,respectively,to minimize the sensing coherence of the system in the form of Frobenius norm.The sampling efficiency of the system and the reconstruction quality for spectral images are dramatically promoted in the optimized dual-disperser CASSI system with RGB image sensor.Moreover,we propose a lens-free single-pixel compressive spectral imaging system based on RGB detectors.In this system,the spatial and spectral modulation are conducted with structured illumination and RGB detectors respectively,which realizes lens-free,low-cost,and stable spectral imaging.The corresponding reconstruction algorithm is developed with simultaneous optimization of the spatial modulation patterns and the over-complete dictionary according to the coherence minimization.Taking full advantage of the intrinsic sparsity of the spectral images in both spectral and spatial dimensions,with the optimized reconstruction algorithm,the spatial modulation patterns and the over-complete dictionary are adapted to each other,which improves the reconstruction quality of the spectral imagesFinally,the insufficiency of the current compressive spectral imaging technologies is discussed,and the prospect of spectral imaging towards high-speed,high-resolution,and low-cost future is provided.