Study On Demosaicking For Hyperspectral Image Based On Low-rank Constraint

Hyperspectral imaging technology capturing spatial and spectral information simultaneously,which has been widely used in biomedicine,geological exploration,agriculture,military and other fields.According to different imaging methods,spectral imaging technology can be divided into scanning spectral imaging and snapshot spectral imaging.The snapshot spectral imaging technology based on spectral filter array(SFA)can obtain the spatial and spectral information through one-time imaging,which has irreplaceable advantages in dynamic scene and the field requiring real-time monitoring.The snapshot spectral imaging technology based on SFA directly acquires a two-dimensional mosaic image of the under-sampling of spatial and spectral information,and the original three-dimensional spectral image needs to be reconstructed by an image demosaicing process.An efficient demosaicing method is critical to imaging quality,which is the main research content of this paper.This paper mainly studies the hyperspectral image demosaicing method.And we proposed a hyperspectral image demosaicing method based on low-rank constraint,the validity of the proposed method is verified by public data sets and experimental data.The main contents of this dissertation are as follows:(1)Study on RGB Image Demosaicing Based on Low-Rank Constraint: Several classical RGB image demosaicing methods are analyzed.The low-rank constraint is applied to the traditional RGB image demosaicing.(2)Study on Hyperspectral Image Demosaicing Based on Low-Rank Constraint:Hyperspectral images are highly correlated in both spatial and spectral dimensions,as hyperspectral data cubes have low-rank properties.The low-rank constraint of image demosaicing can be realized by matrix low rank approximation.The effectiveness of is verified by simulation and experiment,and the effects are evaluated from two dimensions of space and spectrum.(3)Study on Hyperspectral Image Demosaicing Based on Image Fusion: Fully sampled grayscale image is used for information fusion to improve spatial quality of demosaic image.The effectiveness of is verified by simulation and experiment,and the effects are evaluated from two dimensions of space and spectrum.The main innovations of this paper are: 1)According to the low-rank property of hyperspectral image,especially the spectral dimension,a hyperspectral image demosaicing method based on low-rank constraint is proposed.This method has better reconstruction effect in both spatial and spectral dimensions.2)Incorporating grayscale image information into the image demosaicing process effectively improves the spatial reconstruction effect and presents more detailed information.