| Hyperspectral data contain rich spatial and spectral information,which makes it widely used in various fields due to the combination of image and spectral information.In recent years,the demands for higher spatial resolution and higher spectral resolution has been increasing,so the massive amount of hyperspectral data has consumed a large amount of resources in data collection,storage,and transmission process.The theory of compressed sensing provided a new way to address the above problems.A coded aperture snapshot spectral imaging system based on its principle compressed data while collecting data,which greatly improving efficiency.The main research contents and results of this thesis include the following two parts:(1)A coded aperture snapshot spectral imaging system using a spatial dimension encoding scheme was designed and built.This thesis conducted theoretical analysis,optical simulation and target simulation reconstruction to verify the correctness of the scheme.In terms of the selection of key components in the imaging system,the characteristics of spatial modulation devices and spectral modulation devices were analyzed from the reflective and transmissive perspective.Based on the actual situation,glass mask plate and transmission grating were selected to build the coded aperture snapshot spectral imaging system.In order to control the errors of various optical devices,this thesis used laser positioning strategy combined with the principles of light refraction and reflection to build the imaging system,mainly included initial optical axis calibration,lens adjustment,glass mask plate adjustment,transmission grating adjustment,and so on.After the construction of the imaging system was completed,relative radiometric calibration was performed on this system,and a matching algorithm for masks and pixels was designed to perform spatial calibration on the imaging system,common spectral calibration schemes were introduced.Finally,system imaging experiments were conducted.The experimental results showed that the imaging system has achieved compressed sampling of target scene data,which verified the correctness of the imaging system scheme and the reliability of the construction process from a practical perspective.(2)In the practical application of plant leaf hyperspectral data,only the leaf region needs to be focused,while the existing compressed sensing reconstructed algorithm does not distinguish between leaf region and non-leaf region,which results in data redundancy.Aiming at this problem,this thesis proposed a novel recursive sub-tensor hyperspectral compressive sensing of plant leaves based on multiple arbitrary-shape regions of interest(RSTHCS).RSTHCS removed non-leaf regions under the guidance of mask image,only extracted leaf sub-tensors in the leaf regions using the maximum inscribed rectangle algorithm and reconstructed them with tensor compressed sensing algorithm.Experiments were conducted using tea leaf hyperspectral image and soybean leaf hyperspectral image.The experimental results showed that RSTHCS effectively distinguished between leaf regions and non-leaf regions,and still achieved high-quality reconstruction of leaf regions at low sampling rates,while retaining more structural information of hyperspectral data. |
