Research On Reconstruction Algorithm Of Plant Hyperspectral Data Based On Compression Sensing

Hyperspectral data since it is possible to obtain spectral information on a continuous wavelength,can obtain the information that multi-spectral remote sensing cannot express by fine spectral analysis and information extraction.This feature makes the hyperspectral data show significant advantages in plant physiological and biochemical parameter inversion etc.With the development of plant hyperspectral technology,the traditional data compression technology was challenged by a substantial increase in the amount of data acquisition,transmission,data analysis and process.The issue surrounding the establishment of a plant for effective hyperspectral data compression and reconstruction methods,improve data storage,transmission efficiency while maintaining spectral information needs of plant spectral analysis application for a study.The main contents of this paper are as follows:(1)In this paper,the development of hyperspectral technology and the difficulties in the research are introduced.The compressed sensing technology is introduced to the compression and reconstruction of hyperspectral data.Firstly,the compressive perceptual reconstruction of plant physiological and biochemical parameters was studied by using the PROSPECT model data.Then the distributed compression perception oriented to the plant spectral characteristics was proposed and the effectiveness of the algorithm was studied through real data experiment;(2)Through the most widely used PROSPECT model at home and abroad to produce the simulation of plant leaf reflectance data,using different sampling rate of plant spectral compression reconstruction experiment,with moisture content,carotenoid and chlorophyll content of key physiological and biochemical parameters for the inversion of the target.On the basis of the correlation between the spectral data of the plants,the effect and influence of the signal compression reconstruction were discussed in 3 aspects,the original spectrum,the spectral index and the inversion model respectively.In the analysis of the test results,the normalized root mean square error was introduced as the evaluation index,the variation of the original spectral reconstruction error and the sampling rate were studied,and the sensitivity of different spectral indices to the sampling rate was analyzed.Based on the corresponding spectral index and physiological and biochemical parameters,the regression model of partial least squares regression was established.The relationship between the error of reconstruction model and the sampling ratewas studied.The results showed that compression perception could effectively preserve the spectral information of plant;(3)By analyzing the correlation between hyperspectral data bands,a hyperspectral sparse joint model was established,and the compressed hyperspectral data were reconstructed by distributed compression sensing.Experimental results show that the distributed compressed sensing is more efficient than the classical orthogonal matching pursuit(OMP)and the gradient projection reconstruction(GPSR)algorithm,and the peak signal to noise ratio(PSNR)and the efficiency of the algorithm are significantly improved;(4)The distributed compressed sensing based on spectrum adaptive grouping was proposed.by introducing the PSNR as an adaptive grouping threshold,the effect of different threshold grouping and the reconstruction effect of different thresholds were studied,the results show that the adaptive PNSR has significantly improved.According to the characteristics of hyperspectral data in the application,the reconstruction algorithm of some key bands is improved.The improved algorithm can further improve the efficiency of the algorithm in the premise of maintaining the accuracy of reconstruction;In this paper,we study the compressive sensing reconstruction of PROSPECT model data and real hyperspectral data.The results show that the spectral compression and reconstruction method based on the compressive sensing theory can significantly reduce the amount of hyperspectral data and maintain the key information of plant spectrum,which can effectively support the processing and analysis of plant hyperspectral data.