Extraction Of Rice Planting Information Based On Time Series Remote Sensing

Rice is one of the most important crops in food security in Asia.It is an important necessity for humankind and is related to the survival and well-being of many people around the world.The development of rapid and accurate rice planting information extraction technology is of great significance for rice planting monitoring,yield assessment,and production management in Southeast Asia.The Indochina Peninsula has a tropical and subtropical monsoon climate in most areas.The average annual temperature is high and the rainfall is large.It is suitable for rice cultivation.It is one of the important rice production and export areas in the world.Many regions can grow multi-season rice.The rice planting pattern in this region is flexible,there is no fixed seasonal variation characteristics,and the amount of information contained in the single-view image is limited,making it difficult to obtain complex planting patterns.Based on multi-source time series remote sensing data,this paper constructs a rice planting information extraction method based on time-series feature fusion,which extracts rice planting patterns in the small areas of the Chao Phraya River,Thailand,and Cambodia,respectively.It is cloudy and rainy,and the rice planting information in the tropical region is flexible.Extraction provides a new idea.The main research contents and conclusions of this article are as follows:Based on the Sentinel-1 SAR time series data,this paper proposes a method for extracting rice planting information that combines time-series statistical parameters and time-series similarity features,and achieves high-precision extraction of complex rice planting patterns in tropical regions.First,use all available Sentinel-1 SAR data during the year to construct time series curves of backscatter coefficients of different spatial units(pixel-based and object-based)to extract time-series characteristic parameters.Membership between time-series curves on different spatial units and standard backscatter coefficient curves;combining time-series characteristic parameters and membership of time-series curves,supervised classification using random forest classifiers for rice planting patterns in the study area Perform extraction.The results show,the user accuracy of singlecrop rice with object-based method was 81.46%,the producers accuracy was 82.00%,the user accuracy of multi-crop was 88.00%,and the producers accuracy was 84.08%.The accuracy of objectbased classification was better than pixel-based classification.This section uses Cambodia as an example.Based on the MODIS image set,the NDVI time series of 2011 and 2016 are constructed respectively.The DTW algorithm is used to calculate the degree of membership between the time series curve and the ground standard curve.Combined,the random forest model is used for supervised classification,the rice planting area in the study area is calculated,and the rice planting pattern is obtained according to the time when the peak of the time series curve appears.Using MODIS time series data to extract the date,duration and maximum flooded area of the flood in this area,and analyze the spatial and temporal distribution of rice under the influence of the flood.Studies have shown that NDVI time series can also be used to obtain rice planting patterns,and the spatial and temporal distribution of rice plantations will vary with the size of the flood.