Studyon Crop Classification Basedon Red Edge Features Of GF-6 WFV And Sentinel-1/2 Data

Crop identification and classification by remote sensing is the basis for monitoring crop planting area,structure,growth,and estimating crop yield.It is an important part of remote sensing monitoring of agricultural conditions,and is of great significance for government departments to formulate agricultural policies and ensure national food security.With increasing data on medium and high resolution remote sensing satellite at home and abroad,the Gaofen series and the Sentinel series satellites have provided a large number of reliable data sources for crop remote sensing monitoring,and the related red edge features and backscatter features are gradually applied to agricultural remote sensing fields such as crop remote sensing classification.However,for the application of different red edge features and backscattering features of different data sources to the classification of crops and other ground features,there are still some problems such as insufficient mining and analysis of classification features,and insufficient combination between different features.In this study,Hengshui City,Hebei Province was selected as the study area,and the data derived from domestic GF-6 WFV data and ESA Sentinel-1/2 data.Crop classification was carried out by analyzing different features of single temporal GF-6WFV data,time series GF-6 WFV data and time series Sentinel-1 and Sentinel-2 data,and the effects of different red edge and backscattering features from different data sources on crop classification were analyzed.At the same time,a comparative study of red edge features of GF-6 WFV and Sentinel-2 data was conducted to improve the accuracy of crop identification and classification by remote sensing.The purpose of this dissertation is to provide support and reference for the better application of GF-6 WFV data and Sentinel-1/2 data and their related red edge and backscattering features in crop classification by remote sensing.The main research work and achievements of this dissertation are as follows:(1)Crop classification with different red edge features was studied based on a single temporal GF-6 WFV data.Through the analysis and extraction of different red edge features,different classification schemes were studied and designed for crop classification.The results showed that the red edge spectral features,red edge texture features and red edge indices of GF-6 WFV data can improve the accuracy of crop classification in varying degrees,as observed 2.2% and 9.95% for the overall accuracy.The extracted red edge features based on red edge 710 nm band can improve the accuracy of crop classification compared with those based on red edge 750 nm band,which is helpful to promote the popularization and application of domestic GF-6 WFV data and its red edge bands in agricultural remote sensing.(2)For the time series GF-6 WFV data,crop classification based on the combination of traditional NDVI time series and red edge index time series was carried out.By constructing different red edge indices time series and conducting feature selection and importance evaluation,the NDRE red edge index time series was optimized;Combining NDVI time series,seven different lengths of NDVI,NDRE and NDVI&NDRE time series were designed,and the impact of different time series lengths and red edge indices on crop classification was analyzed.The results showed that NDVI time series is more beneficial to improve the overall accuracy of crop classification than NDRE red edge index time series,and the overall accuracy is between 1.48% and11.49%,while NDRE red edge index is more helpful to identify summer crops with similar phenology from August to October,such as summer corn,spring corn and cotton,and can assist NDVI to improve the accuracy of crop identification and classification.The overall accuracy of the NDVI&NDRE time series combined with different time series features is enhanced ranging between 1.11% and 4.82% compared with that of the NDVI time series of different length,which is helpful to improve the accuracy and timeliness of crop classification.(3)Crop classification based on different backscatter features and red edge index of time series Sentinel-1 and Sentinel-2 data was carried out.By combining with the climate characteristics and different feature importance evaluation and actual classification results of Hengshui City in the study area,the classification scheme combining VH backscattering feature and NDre2 red edge index feature of Sentinel-1and Sentinel-2 time series data was designed,and the crop classification and accuracy evaluation with different time series data classification schemes were carried out.The results showed that the overall accuracy of the classification scheme based on the direct combination of VH and NDre2 and the combination of local climate characteristics is11.27% and 9.14% higher than that of VH time series,and 2.46% and 0.33% higher than that of NDre2 time series,respectively.The combination of Sentinel-1 SAR data with Sentinel-2 optical data can make full use of the backscattering and red edge features of different remote sensing data sources,reduce the influence of weather factors such as cloud,rain and fog,improve the accuracy of crop identification and classification,and promote the fusion and popularization of different classification features of different remote sensing data sources.(4)Comparative analysis of crop classification with different red edge characteristics was carried out based on GF-6 WFV and Sentinel-2 data.The single time phase and time series GF-6 WFV and Sentinel-2 data with similar time phase in Hengshui City were selected,and the characteristic differences of different red edge bands of single time phase GF-6 WFV and Sentinel-2 data were discussed by means of correlation analysis,band information quantity and sample separability analysis.By constructing three different red edge index time series of NDRE,NDVIre1 and NDVIre2,combined with different red edge spectra,red edge textures and red edge indices features of single phase GF-6 WFV and Sentinel-2 data and different red edge indices time series features of time series GF-6 WFV and Sentinel-2 data,different red edge feature classification schemes were designed and the classification results were compared and analyzed.The results showed that there is a significant correlation between the red edge and other bands of GF-6 WFV and Sentinel-2 data.Under the premise of the same spatial resolution,classification samples,classification algorithm and similar time phase,except for the red edge 710 nm texture feature of GF-6 WFV data and NDVIre2 time series,the classification accuracy of single time phase and time series red edge feature of Sentinel-2 data is better than that of the corresponding GF-6WFV data.This study promotes the comparative analysis and application of red edge features of different remote sensing data.There are 45 Figures,44 Tables and 271 References in this dissertation.