| 摘要: |
| 【目的】特征光谱指数构建是提高农作物类型识别效率的前提，也是构建通用型识别
方法的基础。【方法】基于2017 年8 月8 日河北省冀州市5 m 空间分辨率的Rapideye 影像，
采用红边、近红外波段反射率之和构建了棉花提取指数（Cotton Extraction Index，CEI），结
合同期水体、裸地（含城镇建筑）掩模处理，实现了棉花类型识别。CEI 指数方法具体过程
是：样本点选择、水体及裸地（含城镇建筑）掩模处理、CEI 指数构建、棉花识别及精度
验证。其中，样本点采用随机方法获取，共获取了5 144 个样本点，作为最佳CEI 阈值以
及精度验证的依据。【结果】基于CEI 方法提取结果的总体精度达到88.80%，Kappa 系数达
到0.751 7。为了评价CEI 指数方法的相对精度，分别采用最大似然分类方法和随机森林分
类方法对影像进行分类和精度验证。结果表明基于最大似然分类方法的棉花提取总体精度和
Kappa 系数分别为86.53% 和0.698 3，采用随机森林方法的棉花提取总体精度和Kappa 系数
分别为90.12% 和0.766 7。【结论】对比3 类方法，基于CEI 指数方法获得的棉花提取结果，
与其他2 种常用方法的提取精度相当。该方法具有简单、高效，对样本依赖性较小的优势，
在时效性要求较高的农作物面积识别中具有较高的应用价值。 |
| 关键词: 棉花 遥感识别 Rapideye 特征光谱指数 随机森林 最大似然分类 |
| DOI：10.12105/j.issn.1672-0423.20190503 |
| 分类号: |
| 基金项目:中国农业科学院创新工程领军人才C 类（农业遥感） |
|
| Construction and application of cotton characteristic spectral index based on Rapideye image |
|
Wang Limin, Liu Jia※, Yao Baomin, Gao Jianmeng, Ji Fuhua
|
|
Institute of Agricultural Resources and Regional Planning，Chinese Academy of Agricultural Sciences，Beijing 100081，China
|
| Abstract: |
| ［Purpose］Construction of characteristic spectral index is the key to improve the extraction efficiency of agricultural crops，and it is also the basis of developing identification methods which can be commonly used［. Method］Cotton Extraction Index（ CEI） was constructed by summing up the reflectance values in red-edge and near-infrared band of 5m resolution Rapideye image which covered Jizhou city in Hebei province and was acquired on 8 August 2017. Cotton area was identified with the CEI method after mask processing which aimed to eliminate the waterbody and bare land（ including buildings）. The whole process of CEI method consists of sample selection，mask treatment of water body and bare land，calculation of CEI， cotton identification and accuracy assessment. The threshold of CEI to separate cotton and other objects is considered to be the most important figure，which is obtained by 5 144 sample points. ［Result］The overall accuracy and Kappa coefficient of the classification result based on CEI method were 88.80% and 0.751 7 respectively. In order to evaluate the effectiveness of CEI method，random forest（ RF） and maximum likelihood classification（ MLC） method were used. The result showed that the overall accuracy and Kappa coefficient of the result acquired by MLC were 86.53% and 0.698 3. The figures for RF were 90.12% and 0.766 7 respectively. ［Conclusion］The cotton extraction accuracy based on CEI method was similar to those by MLC and RF methods and the CEI method was simpler，more efficient and less dependent on samples， which was applicable to the instant crop extraction. |
| Key words: cotton remote sensing identification Rapideye characteristic spectrum index random forest maximum likelihoods |
