Stripe Noise Removal For Lunar Hyperspectral Imagery Employing Low-rank Framework

Hyperspectral imagery(HSI)is one of emerging tools to explore the physical properties and chemical composition of the lunar surface.Due to the limitations of sensor manufacturing and disturbances from the external environment,there are inconsistent responses within the pushbroom linear charge-coupled device arrays of the lunar hyperspectral imaging sensors,resulting in serious stripe noises on the obtained images.Moon Mineralogy Mapper(M~3)is the most commonly used lunar HSI dataset with the widest coverage and excellent resolutions;however,dense and nonperiodic stripes distributed across all bands in M~3images hinder visual interpretation as well as their use in subsequent applications.The stripes on the lunar HSI are complicated with various intensities and spatial distri-butions,and they are so dense that they destroy the high correlation and redundancy inside the hyperspectral data.However,most existing algorithms cannot handle this case well for the lack of generalization.Therefore,this dissertation aims at developing an efficient and high-performance destriping model based on the flexible and rigorous low-rank variational framework.The main contents are shown below:(1)A robust stripe processing model for dense and irregular stripes is proposed.Based on the phenomenon that the mean-cross track curves of the degraded and clean images have the same trend but different fluctuations,this model embeds Hodrick-Prescott decomposi-tion into the low-rank matrix restoration framework to recover the severely degraded images.Simulated and real experiments conducted on typical regions on the Moon with various levels of corruption demonstrate that the proposed model rapidly achieves favorable performance against state-of-the-art approaches.(2)A stripe removal method with high information fidelity is developed to handle the weakness of the method proposed in(1).On the foundation of hyperspectral imagery sub-space identification theory,this model divides the hyperspectral image into the spatial princi-pal component and the spectral orthogonal basis.With the application of the weighted nuclear norm,this method is adaptive to the stripe noise level.The experiments and integrated band depth mapping results show that this method has better overall performance and higher infor-mation fidelity than the algorithm proposed in(1).Additionally,the extended experiments on terrestrial observations further verify its great generalization.