Research On Super-resolution Reconstruction Methods For Multispectral Imaging

Spectral imaging technology has an urgent need in the fields of military investigation,medical treatment,satellite sensing,weather prediction,disaster prevention and so on.Especially with the increasing demand for earth observation and outer space exploration,the demand for spatial and spectral resolution of spectral images is becoming higher and higher.The traditional spectral imaging technique is to obtain three dimensional spectral data based on two dimensional detection sensors,and the imaging quality depends heavily on the performance of the sensor.However,the sensor density is restricted by the manufacturing processes,volume,power consumption,transmission bandwidth,cost and other factors.In2006,the development of compressed sensing theory(CS)has brought innovation and theoretical support to computational imaging technology,and the framework of computational imaging technology based on compressed sensing theory has also been put forward.In the framework of computational imaging technology,the imaging process is divided into two divisions: the observation process and the data restoration process.The observation process mainly obtains the measurements by coding and sampling of the target scene.The data restoration process can be achieved by solving optimization problems based on the prior knowledge of the scene.The key point is to establish a suitable optimal solution model to obtain the high-resolution spectral image.For multi-spectral imaging system,there are some drawbacks in existing super-resolution methods,due to jitter in the hardware or load platform during imaging,noise interference,system defocus,etc.In response to these issues,this paper starts with the principle and process of imaging and focuses on super-resolution methods for multi-spectral imaging based on coded sensing computational spectral imaging framework.The purpose of this study is to improve the spatial resolution and spectral accuracy,and thus improves the quality of the reconstructed multi-spectral image.The main contributions and innovation points of the thesis are as follows:1?This paper studies the frame of computational spectral imaging based on compressive sensing and focuses on the optimization inversion stage in this framework.The physical imaging model is used to analyze the reflection of natural light on the surface of the object to simulate the natural scene imaging process.We use a physically-induced model to explore spatial-spectral sparsities of spectral scenes and infer spatial alignment property of spectral Laplacians.Then,we use these prior knowledge to establish the second order Laplace super resolution model and use the super resolution model for color image demosaicking.The experiments show that our method can effectively improve the edge color accuracy and antinoise ability of color images,which validates the effectiveness of spatial-spectral sparsity analysis proposed in this paper.2?Aiming at the problems of the low spatial resolution and poor spectral accuracy of spectral imaging in CASSI system,we choose CASSI system to obtain the low resolution spectral scene.Several experiments are conducted to evaluate the performance of the proposed method in comparison with traditional super-resolution methods,such as A+ method,BSSC method,NCSR method.The simulation results show that our method has an average increase of 0.3dB in peak signal-to-noise ratio PNSR,and it is also better than traditional methods in terms of structure correlation.More importantly,the existing method obtains false color speckle noises near edges,whereas proposed method does not.Furtherly,some experiments are conducted to validate the advantages of the proposed method in improving the spectral accuracy.Simulation results demonstrate the efficacy of the proposed method.