Simulation Of Multispectral Images To Hyperspectral Images Based On Spatial Vector Analysi

Hyperspectral remote sensing technology can obtain finer and richer spectral information of ground objects.However,due to the limitations of hyperspectral loads and atmospheric influences,the coverage and timeliness of hyperspectral data are insufficient.In response to this issue,researchers have conducted a large number of hyperspectral image simulation studies.Among them,although the hyperspectral image simulation method based on the standard spectral library has important significance in the design and development of hyperspectral loads,in real-world hyperspectral applications,the standard spectral library data has differences in imaging spectra and richness,which cannot reflect the spectral characteristics of the same category of ground objects in different states and phenology.In addition,traditional classification based retrieval methods have low efficiency,so there is still room for improvement in the accuracy and efficiency of hyperspectral image simulation for real-world hyperspectral applications.Therefore,in order to improve the simulation accuracy and efficiency of hyperspectral images,this paper proposes a hyperspectral image simulation method based on scene spectral library.This method mainly includes three parts:the construction of scene spectral library,spectral retrieval,and spectral matching.In terms of spectral library construction,compared to standard spectral libraries,this article constructs a scene spectral library that can reflect the spectral characteristics of the same category of ground objects in different states and phenology in actual imaging scenarios.In terms of spectral retrieval,this paper proposes a spectral retrieval method that integrates category,NDVI,and time labels,in response to traditional category based retrieval methods.In terms of spectral matching,space vector analysis is used to match spectra.This method not only considers the angle between spectral vectors,but also considers the consistency of spectral vector modulus.In response to the method proposed in this article,Dezhou City,Shandong Province was selected as the research area for experimental verification.High score one(GF-1)WFV multispectral data was used to simulate high score five(GF-5)AHSI hyperspectral data,and the simulation results of the three spectral retrieval methods were compared with real hyperspectral images.The experimental results showed that the simulation results of the three spectral retrieval methods perform well,among which the simulation effect based on time tag retrieval was the best,with an average R~2 of 0.64,a maximum of 0.76,an average RMSE of0.033,and a minimum of 0.021.The spectral angular distance between the simulated hyperspectral and real hyperspectral spectra of typical terrain in the study area was above98%,with a maximum of 99.5%.In addition,in terms of operational efficiency,this method greatly shortened the simulation time and improved the simulation speed by 88%.At the same time,the simulation results were compared with those of simulation methods based on standard spectral libraries.The average R~2 increased by 0.3,the average RMSE decreased by0.04,and the spectral angular distance of typical features in the study area was also improved.