基于Sentinel数据的新疆奇台县小麦作物识别和秸秆产量估算

Wheat Crop Identification and Straw Yield Estimation in Qitai County Based on Sentinel Data

【目的】新疆奇台县小麦作物种植面积大,秸秆资源丰富。监测并获取秸秆资源的分布与产量,对于促进秸秆综合利用、减少环境污染具有重要意义。【方法】以奇台县境内小麦作物为研究对象,融合多时相的Sentinel-2光学和Sentinel-1雷达数据,选择合适的特征,使用三种机器学习方法对奇台县小麦作物进行识别和小麦秸秆估产。【结果】使用随机森林分类器的分类结果最好,在仅使用Sentinel-1的雷达极化特征分类时,总体精度为96%。在使用Sentinel-1的雷达极化特征与Sentinel-2光学波段特征分类时,总体精度为94%。当使用Sentinel-1的雷达极化特征、Sentinel-2光学波段特征以及植被特征指数的组合分类时精度最高,总体精度可达到98%,Kappa系数可达到97%。基于地面抽样调查数据,估算小麦作物秸秆每公顷产量为27007 kg,总产量为100615.5 t。【结论】在小麦抽穗期,特征组为Sentinel-1的雷达极化特征、Sentinel-2光学波段特征以及植被特征指数的组合,使用随机森林分类效果最好。

【Objective】The wheat crop growing area in Qitai County is extensive,and straw resources are abundant.In order to improve straw consumption and prevent environmental pollution,it is important to detect and obtain the distribution and yield of straw resources.【Methods】This study takes the wheat crop in Qitai County as the research object,integrates the multi-temporal Sentinel-2 optical and Sentinel-1 radar data,selects the relevant features,and adopts three machine learning approaches to identify the wheat crop and estimate the wheat straw yield in Qitai.【Results】The results show that the random forest classifier produced the best classification results,with an overall accuracy of 96%when using only Sentinel-1 radar polarization data for classification.When the combination of Sentinel-1 and Sentinel-2 optical band radar polarisation data was used for classification,the overall accuracy was 94%.With an overall accuracy of 98%and a Kappa coefficient of 97%,the combination of radar polarisation information from Sentinel-1,Sentinel-2 optical band,and vegetation index produced the highest classification accuracy.The predicted production of wheat crop straw was 2700 kg per hectare,and the totalproduction was 100616 t based on the ground survey sample data.【Conclusion】During the wheat tassel stage,the random forest classifier with the combination of rader polarisation information from Sentinel-1,Sentinel-2 optical bands,and vegetation index produced the highest classification accuracy.