Remote Sensing Image Fusion Algorithm For Multi-source Remote Sensing Data

In recent years,with the development of remote sensing technology,numerous new satellite sensors have emerged.The remote sensing images obtained by different remote sensing sensors have their own advantages and disadvantages due to the different imaging mechanisms between the sensors.In order to fully understand the characteristics of objects and their changes,multi-source satellite sensor data fusion methods were invented,that is,the technology of fusing image data obtained by different types of sensors.The resulting fused image can take advantage of different satellite sensor data to cope with more and more demanding remote sensing applications.Optical spatiotemporal fusion and the fusion of SAR(Synthetic Aperture Radar)and optical remote sensing images play important role in this aspect.In view of this,this paper studies optical remote sensing image spatio-temporal fusion algorithms,optical and SAR image fusion methods and their application fields,and expands new directions for multi-source remote sensing image fusion and its applications.This paper divides the experimental process into two parts,one is the CASA(Carnegie-Ames-Stanford Approach)model NPP(Net Primary Productivity)simulation based on optical space-time fusion;the other is object-oriented ground object classification based on SAR and optical fusion.The specific research contents and results are as follows:(1)Based on the classical spatiotemporal fusion methods FSDAF(Flexible Spatiotemporal Data Fusion),ESTARFM(Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model)and STARFM(Spatial and Temporal Adaptive Reflectance Fusion Model),a 30m resolution NDVI(Normalized Difference Vegetation Index)was constructed vegetation index data.NDVI reconstruction results of threee methods were used for linear regression analysis with MODIS reference data,and the pearson correlation coefficients were:0.83,0.77,and 0.76,respectively.Simulation of NPP were conducted using the constructed NDVI dataset and the accuracy evaluation shows that:under the highly heterogeneous usage scenarios,the targeted optimization spatiotemporal fusion methods such as FSDAF and ESTARFM have better consistency than the STARFM method.The results of FSDAF has the smallest gap with the true value.The simulated NPP value of ESTARFM is relatively low,but the trend of NPP is consistent with the true value.(2)An improved NSST-MSMG-PCNN method is proposed to realize the fusion of optical and SAR images.The NSST-PCNN(Nonsubsampled Shearlet Transform-Pulse Coupled Neural Network)method in the field of image fusion is improved,and the MSMG(Multi-Scale Morphological Gradient)morphological operator is introduced to detect edge information.In this paper,,combined with Brovey transform,IHS transform,PCA transform,NSCT transform,NSST-PAPCNN transform and NSST-MSMG-PCNN transform,the fusion results of polarizations of VV and VH are analyzed respectively through the evaluation criteria of image fusion accuracy and visual interpretation.The results show that NSST-MSMG-PCNN presented has better fusion effect,and VV polarization fusion can achieve better results in the two study areas.The correlation coefficients reach 0.9193,0.9012,respectively,while the spectral distortion are 27.2153 and 24.0015.(3)Land object classification based on fusion results of optical and SAR images.The fusion results of the NSST-MSMG-PCNN method in this paper have improved accuracy compared to other fusion methods and the direct use of source images for classification.In the rural application scenario,the overall accuracy and Kappa coefficient are 0.8678 and 0.8137,respectively,which are 8.24%and 11.56%higher than the Landsat-8 source image classification;In the urban application scenario,the overall accuracy and Kappa coefficient are 0.88 and 0.7644,which are 23.60%and 42.93%higher than the direct use of Landsat-8 source image classification.