Abstract: In order to compare differences between Sentinel-2and GF-6 WFV imagery in crops identification,based on Sentinel-2 and GF-6 WFV remote sensing data,this study used K nearest neighbor and maximum likelihood classification methods to extract peanut planting area in Yulin Township,Xuchang City,and constructed confusion matrix by ground sample points for accuracy verification.The relative errors of the extracted peanut planting area were compared based on the measured data.The results showed that two classification methods were effective in extracting peanut planting area from two data sources and could meet the actual needs. The mapping accuracy was above 85%,the user accuracy was above 80%,and the relative error was within 10%. Peanut planting area was mainly concentrated in the northwest and southeast regions,and there were a few sporadic distributions in the northeast and southwest regions.By comparison,the object-oriented K nearest neighbor method could better avoid pixel mis-segmentation and leakage-segmentation problems in complex terrain area.K nearest neighbor method was superior to traditional pixel-based maximum likelihood classification in terms of overall accuracy,Kappa coefficient,peanut planting area mapping accuracy,user accuracy,and relative error.In terms of different classification methods of the same data source,the classification accuracy of two data sources using K nearest neighbor method was higher than the maximum likelihood classification.It showed that compared to the pixel-based classification method,K nearest neighbor method could make full use of the spectrum and texture feature,and obtain higher extraction accuracy. In terms of the same method and different data sources,the extraction accuracy of the peanut planting area based on Sentinel-2 by the maximum likelihood method was lower than that of GF-6 WFV,and the extraction accuracy of the peanut planting area based on Sentinel-2 by the K neighbor method was higher than that of GF-6 WFV.Because the spatial resolution of Sentinel-2 data is higher than that of GF-6 WFV,and the detail expression is better,it is more suitable to extract small-scale areas with complex planting structures.

Key words: Sentinel-2, GF-6 WFV, Maximum likelihood method, K nearest neighbor, Peanut, Area extraction, Difference analysis

摘要： 为比较基于Sentinel-2与GF-6 WFV数据对花生种植面积提取的差异，以许昌市榆林乡为研究区域，选取Sentinel-2和GF-6 WFV遥感数据相同波段，采用K邻近法和最大似然法提取花生种植面积，基于样本点构建混淆矩阵进行精度验证，并结合地面样方实测数据，比较对花生种植面积提取的相对误差。结果表明，2种分类方法对2种数据源的花生种植面积提取效果均可满足实际需要，制图精度均大于85%，用户精度均大于80%，相对误差均小于10%。提取结果显示,花生种植面积空间分布主要集中于西北部和东南部，东北部和西南部面积较少且分布零星。通过对比发现，采用面向对象的K邻近法能更好地避免复杂地物类型区像元错分及漏分问题，其总体精度、Kappa系数以及花生种植面积的制图精度、用户精度和面积相对误差等参数均优于传统基于像元的最大似然法。在同数据源不同分类方法时，2种数据源利用K邻近法的分类精度均高于最大似然法。说明面向对象的K邻近法可充分利用地物光谱特征及纹理特征，比基于像元的最大似然法取得更高精度的提取结果。在同方法不同数据源时，利用最大似然法对Sentinel-2花生种植面积的提取精度低于GF-6 WFV，利用K邻近法对Sentinel-2花生种植面积的提取精度高于GF-6 WFV。综上，由于Sentinel-2 10 m融合数据的空间分辨率高于16 m的GF-6 WFV，对细节的表达效果更佳，更适合用于提取种植结构复杂的小尺度区域。

关键词: Sentinel-2, GF-6 WFV, 最大似然法, K邻近法, 花生, 面积提取, 差异分析