多时相Sentinel-2A与SPOT-7影像在油菜识别中的差异

[目的]研究多时相Sentinel-2A识别种植结构复杂的小尺度区域中的油菜面积，获取高精度的作物分布信息。[方法]以多时相Sentinel-2A和一景SPOT-7数据为数据源，选取种植结构复杂的小尺度区农业区为研究区，构建不同特征向量组合，利用支持向量机（support vector machine，SVM）的方法提取油菜种植面积。[结果]通过对比分析基于不同特征向量组合的油菜识别精度，利用一景油菜最佳识别期内的Sentinel-2A影像可以得到高达89.1%的制图精度和92.1%的用户精度；添加油菜最佳纹理特征后，多时相Sentinel-2A数据的制图精度与用户精度分别提高了2.9%和2.5%，仅比SPOT-7影像的识别精度低了1.7%和2.1%，2种数据的油菜提取精度差异进一步减小；Sentinel-2A与SPOT-7数据油菜最优分类结果对比后，一致性精度和Kappa系数分别为93.3%和0.89。[结论]多时相Sentinel-2A数据可以很好地识别种植结构复杂地区的油菜，加入最佳纹理信息能够提高油菜的识别精度；Sentinel-2A可以广泛应用于小尺度区域作物分布信息的快速提取。
[Objectives]The study is to explore the ability of Sentinel-2A to identify rape in complex planting areas and the possibility to extract rape with the same high accuracy as high spatial resolution images.[Methods]The study site was the Gaochun District of Nanjing,the capital of Jiangsu Province,China. Multi-temporal Sentinel-2A and one SPOT-7 images were obtained during the flowering stage of rape. Different combinations of spectral and texture information of two images were classified to map rape by using support vector machine,and then classification accuracy achieved using different combinations were evaluated.[Results]When using spectral information only,the producer’s and user’s accuracy of Sentinel-2A were 89.1% and 92.1%,respectively. After adding the best sensitive texture characteristics of rape,the producer’s and user’s accuracy of Multi-temporal Sentinel-2A data increased by 2.9% and 2.5% respectively,which were 1.7% and 2.1% lower than the accuracy of SPOT-7 respectively. By resampling the optimal rape classification results of SPOT-7 to 10 m,and then comparing them with Sentinel-2A,the consistency and Kappa efficiency were 93.3% and 0.89,respectively.[Conclusions]Sentinel-2A data can identify rape in areas with complex planting structure,and the recognition accuracy can be improved after adding the optimum texture information,and Sentinel-2A can replace high resolution image to extract crop area with high accuracy