Research On Wetland Vegetation Inversion Based On Multi-source Remote Sensing Data

The wetlands at the mouth of the Yellow River are located at the intersection of sea,river and land,and are rich in vegetation species,such as reeds,tamarisk and alkali ponies.However,due to the influence of seawater erosion,sedimentation and other factors,the ecosystem in this area changes more frequently.Therefore,it is extremely important to use remote sensing to extract information on wetland vegetation in the Yellow River estuary and to understand the spatial distribution patterns of different types of wetlands for the rational use of wetland resources and the protection of biodiversity.Efficient coordination and intelligent interpretation of multi-source remote sensing is an important direction of remote sensing development at home and abroad,and various countries have carried out multi-source remote sensing cooperative wetland monitoring projects one after another.In this study,we select multispectral data and hyperspectral data as data sources,combine10 m spatial resolution Sentinel-2 multispectral remote sensing images with 30 m spatial resolution ZY1-2D hyperspectral images to obtain richer feature information,improve the spatial resolution while ensuring the spectral resolution,and conduct an inversion study of vegetation communities in the Yellow River Delta wetlands.The inversion study of vegetation communities in the Yellow River Delta was conducted to explore the effect of different data on vegetation inversion in the study area.The main research and results of the paper are as follows:(1)In the study of vegetation inversion based on multispectral data,an image level spectral index monthly mean time series reconstruction model based on Whittaker smoothing method was proposed for inversion study.The results showed that the overall accuracy of the inversion based on the reconstructed spectral index time series classification method reached 92.07%,and the Kappa coefficient reached 0.91,which was 6.02% better than the overall accuracy of the original spectral index time series inversion before reconstruction.(2)In the study of vegetation inversion based on hyperspectral data,the overall inversion accuracy was relatively low,80.15%,and the Kappa coefficient was 0.77.The spectral characteristics of different vegetation were analyzed,and envelope removal was used to increase the spectral differences of each vegetation.Contrast analysis extracts the intersection of the characteristic spectral ranges among each wetland vegetation as the characteristic spectral bands in which this vegetation differs from the spectra of other vegetation,and the differentiability among vegetation is enhanced.(3)In the study of vegetation inversion based on fused data,G-S transform,PCA and HA methods are used for image fusion processing,CC,STD and PSNR are used as metrics for quantitative evaluation of different fusion methods,and the fusion data after G-S transformation is selected as the data source according to the evaluation results for vegetation spectral extraction and analysis.Random forest method was used to obtain the inversion results,the overall accuracy was 92.47%,and the Kappa coefficient was 0.91.(4)Comparing and analyzing the inversion results based on different data,in general,the inversion accuracy based on fused data is the highest,followed by multispectral data and hyperspectral data.In terms of the classification accuracy of a single land type,the best inversions were obtained using multispectral data for spartina alterniflora wetlands,suaeda wetlands,and tidal flats,the best inversions were obtained using fused data for reed wetlands,clear water,and turbid water,and the best inversions were obtained using hyperspectral data for tamarix wetlands.