Method Of Spatial Distribution Extraction Based On Time-series Sentinel-2 Images For Winter Wheat

Accurate and rapid acquisition of the planting area and spatial distribution of winter wheat can provide important supporting data for winter wheat growth monitoring and yield estimation.Sentinel-2 image has been widely used in the field of agricultural remote sensing due to its advantages of high spatial and temporal resolution and rich spectral information,especially the three red-edge bands it contains are very effective for vegetation health monitoring and have unique advantages in feature identification,thus it has been widely used in the field of agricultural remote sensing.In view of the existing remote sensing image time series extraction methods that mainly use a single vegetation index and a univariate time series similarity judgment method,the extraction accuracy of winter wheat is limited.And the application of red edge vegetation index in crop extraction needs to be explored.Therefore,this paper takes Sentinel-2 images as the data source,selects Huaibin County of Xinyang City as the research area,carries out the research on remote sensing identification of winter wheat based on red-edge band vegetation indices,and constructs a spatial distribution extraction method of winter wheat based on NDVI₇₀₅,NDVI₁ and RENDVI.（1）The quality of sample points is the key to the determination of reference time series and threshold.Based on iterative processing,this paper introduces DTW method to optimize the sample points,extracts each vegetation index according to the sample points collected in the field,then draws the initial reference time series curve based on vector average method,eliminates the abnormal sample points that exceed twice the standard deviation from the DTW of the curve,and iterates repeatedly until the optimal set of sample points is obtained,The results show that this method effectively weakens the influence of phenological differences of different vegetation and mixed pixels.（2）In order to explore the role of vegetation indices constructed by Sentinel-2red-edge band in crop spatial distribution extraction,this paper visually evaluates the separability of different features based on RENDVI,NDVI₇₀₅,NDVI₁,NDVI₂,MTCI,MCARI and REP indices of sample points,and eliminates REP indices with poor separability;Then,four methods,namely,the pinch Cosine of single index time series,Euclidean distance,Manhattan distance and Dynamic Time Warping,are used to measure the similarity of the retained other indices,and then set thresholds to extract the spatial distribution of winter wheat.According to the results,it is found that the extraction effect according to NDVI₇₀₅,NDVI₁,RENDVI is relatively better.（3）In order to improve the similarity measurement method of single index time series,taking time series NDVI₇₀₅,NDVI₁ and RENDVI as characteristic indexes,the similarity measurement methods of angle Cosine,European distance,Manhattan distance and Dynamic Time Warping of multi index time series are constructed by calculating the sum of the distance between the vegetation index to be identified vector and the reference vector at different time sequence nodes.The corresponding threshold is set to extract the spatial distribution of Winter Wheat in Huaibin County,Henan Province in 2020.The results show that the Dynamic Time Warping method of multi exponential time series has the best extraction result,with an accuracy of 93.90%,which is 3.84%,0.82%and 1.44%higher than the angle Cosine,European distance and Manhattan distance of multi exponential time series,and overcomes the problem of unequal time series caused by missing or abnormal pixel points,it provides a new idea for extracting the spatial distribution of winter wheat.