基于多源遥感数据的罗平油菜种植面积提取方法研究

The Research on Identification Methods of Oilseed Rape Based on Multi-source Remote Sensing Data in Luoping County

基于GF-1 WFV和Landsat8 OLI遥感数据,采用面向对象的最邻近法和基于像元的最大似然法,提取罗平县油菜种植区域,并基于实地样本点构建混淆矩阵进行精度验证,比较提取的油菜种植面积的相对误差。结果表明:针对2种数据源,2种方法提取效果均较好。通过对比,采用面向对象的分类方法能更好地避免复杂山区混合像元错分及漏分问题,其在总体精度、Kappa系数以及油菜的生产者精度和用户精度等方面,均优于传统基于像元的分类方法,更适用于喀斯特山区的地物信息提取。在相同数据不同方法方面,采用最邻近法提取GF-1 WFV数据,所得油菜种植面积的相对误差仅为0.74%,精度远高于最大似然法提取该数据的相对误差(-7.37%);对于Landsat8 OLI数据,最邻近法提取的面积精度同样高于最大似然法。相同方法不同数据方面,最邻近法更适用于空间分辨率较高的GF-1 WFV数据,而最大似然法更适用于光谱信息更丰富的Landsat8 OLI数据。

Based on GF-1 WFV and Landsat8 OL1 remote sensing data,the planting area of Brassica campestris in Luoping was extracted by using the nearest neighbor classification and the maxium likelihood method. Based on the actual sampling points,a confusion matrix was constructed to verify the accuracy and the relative error of the planting area of rapeseed was compared. The results showed that whether GF-1 WFV or Landsat8 OL1,both methods had better extraction results. The object-oriented classification method can better avoid the problem of misclassi- fication and leakage of mixed pixels in complex mountainous areas. 1t was superior to traditional pixel-based classification in terms of overall accuracy,Kappa coefficient,rapeseed producer accuracy and user accuracy,and it was more suitable for extracting feature information in the Karst Mountain area. Extracting the same data by different methods,the relative error of rape planting area was only 0. 74% when using the nearest neighbour method to extract GF-1 WFV data, the accuracy was much higher than -7. 37% of maximum likelihood method. For Landsat8 OL1 data,the nearest neighbor method was also more accurate than the maximum likelihood method. Extracting the different data by same methods, the nearest neighbor method was more suitable for GF-1 WFV data with high spatial reso lution, the maximum likelihood method was more suitable lor Landsat8 OLI data with richer spectral information.