基于遥感大数据云计算平台下的水稻识别研究

Rice Identification Based on Remote Sensing Big Data Cloud Computing

传统水稻种植面积估算基于地面测量,再逐级上报,时效性和准确性难以达到监管需求。文章利用遥感大数据云计算平台(Google Earth Engine,GEE),结合多源遥感影像(Sentinel系列,Landsat系列等),对整个湖南省大尺度范围进行了水稻识别算法实验。首先,针对云雾等客观因素造成数据缺失的现象,通过构建一种多源遥感影像数据融合算法,生成了完整的水稻生长周期遥感影像;然后,根据不同作物的生长物候特征,结合归一化植被指数时间序列特征进行分析,对早稻、中稻、晚稻等不同水稻类型进行了有效识别;最后,在精度验证方面,创新性地使用无人机辅助调查的方式对整体实验结果进行精度评价。最终结果表明:早稻、中稻、晚稻的准确率分别为85.15%、89.78%、86.01%,文章算法具有较高精度与鲁棒性,也能够为其他作物识别研究提供一定意义的借鉴与参考。

The traditional estimation of rice planting area is based on ground measurement and then reported step by step,and the timeliness and accuracy are difficult to meet the regulatory requirements.This article proposed a new paddy rice mapping method in large scale based on Earth Engine—A new remote sensing big data cloud computing platform,which was developed by Google Group.Firstly,aiming at objective conditions such as data loss caused by cloudy and fog,a multi-source remote sensing image reconstruction algorithm was constructed to reconstruct the rice growth cycle to obtain a cloudless synthetic image.Then,early rice,middle season rice,late rice were identified based on phenological feature,which was analysed of NDVI time series characteristics of different crops,the result has a well visualization in any web client,Finally,the accuracy of the experimental results was evaluated by UAV.The overall accuracy of early rice,middle season rice,late rice are 85.15%,89.78%,86.01%.it indicate that our result is robust and feasible.