【摘要】 以冬小麦作为研究对象,基于2016年10月-2017年7月共12景GF-1号数据,通过计算不同特征波段构建了多特征冬小麦种植区识别方法。首先计算影像增强植被指数(EVI);再通过样本分离度选择关键期影像,对关键期影像进行主成分变换并计算其纹理特征数据;然后分离原始影像的红波段、近红外波段构建单波段时序数据,以建立包括多特征的数据集;最后基于不同特征组合数据,利用随机森林方法分别提取研究区内冬小麦分布情况。精度评价表明,影像主成分、EVI时序以及红波段、近红外波段特征组合的精度最高可达97.85%,较单独使用时序数据结果提高了约2.5%,说明红波段、近红外波段在作物提取中对精度提升具有一定的贡献。更多还原

【Abstract】 In this paper, taking winter wheat as the research object, based on the GF-1 data of 12 scenes from October 2016 to July 2017, we constructed a multifeature winter wheat planting area identification method by calculating different characteristic bands. Firstly, we calculated the image enhancement vegetation index(EVI), and selected the critical period images by sample separation. And then, we performed the principal component transformation on the critical period images, and calculated the texture feature data. In addition, we separated the red band and the near-infrared band of the original images to construct single-band time series data, and created a data set that includes multiple features. Finally, based on the random forest method, we extracted the distribution of winter wheat in the study area by different feature combination data. The accuracy evaluation result shows that the accuracy of principal component, EVI time series, red band and near-infrared band are up to 97.85 %, which are about 2.5 % higher than the time series data alone. This result indicates that the red band and near-infrared band has certain accuracy in crop extraction.更多还原