Research On Remote Sensing Information Extraction Of Typical Crops In Arid Areas Based On Cloud Computing Platform

China is a large agricultural country.Grain yield is a key factor affecting agricultural development,and crop identification is the initial stage of ensuring global food security.With the global climate drying and warming,the growth cycle of crops has changed and also caused a lot of crop disasters.Nevertheless,large-scale identification of crops can adjust the growth structure and planting system of crops in time.Therefore,whether it is food security or climate change,efficient and timely identification of large areas of crops is particularly important.In this paper,research on multi-feature remote sensing information extraction of typical crops based on Google Earth Engine cloud platform was carried out,sentinel-2B,landsat-8 and sample data were used to extract phenological features and texture features of typical crops,which the extraction targets are corn,wheat and sunflower,the research area is tumed right county of Inner Mongolia autonomous region.Firstly,Sentinel-2B,Landsat-8 remote sensing images and field samples were used the main raw data,and the phenological characteristics of the three crops in 2018 were analyzed by combining the local crop phenology and temperature changes in previous years.The NDVI curves of the aforementioned three crops were generated according to landsat-8 remote sensing images,and the phenological characteristics of typical crops were analyzed.Secondly,the time points which have obvious changes among the aforementioned three curves and the planting and harvesting time of crops are taken as the data sources of multiphase images to analyze the data sources.Thirdly,five temporals were selected.Sentinel-2B is used to analyze eight texture features of remote sensing images in selected temporals,and 40 texture features were generated,and the typical crop texture features were analyzed.Finally,the aforementioned 45 features were analyzed by principal component analysis(PCA),and the first 15 different feature combinations after executing PCA were classified by minimum classification,maximum likelihood classification,decision tree classification and random forest classification.The experimental results show that(i)the classification accuracy based on single feature(i.e.,50.21%~91.26%)is lower than classification accuracy of multiple features(i.e.,55.26%~93.34%)and the classification accuracy based on all features(i.e.,55.26%~93.34%)is lower than that after feature selection(i.e.,62.44%~97.37%).(ii)Different classification methods should combine different feature combinations to achieve the best classification results.(iii)Maize was mainly distributed in three Fangxiang and Haizi townships,with an extraction area of 912.4km2,the wheat mainly distributes in the north of Mingsha Nao Township,Meidai Zhaozhen Township and G6 Beijing-Tibet Expressway,with an extraction area of 110.46km2 and sunflower mainly distributes in Mao Dai Township and Yao Jun Township,with the extraction area of 228.86km2.