基于GF-1 WFV影像的作物面积提取方法研究

黑龙江省是我国粮食生产大省，及时有效地获取黑龙江省的农作物种植面积对后续研究的开展具有重要意义。以黑龙江省五九七农场为例，利用2014年8月30日GF-1卫星16 m空间分辨率影像，通过计算不同特征波段，构建了多特征水稻、玉米种植区识别方法。首先计算影像归一化差分植被指数（NDVI），并将原影像进行主成分变换，以此为基础建立包含多特征的数据集。然后利用不同地物类型之间在各特征波段的差异，基于CART算法构建决策树，分别提取研究区内的水稻和玉米。精度评价结果表明，分类的总体精度达到96.15%，Kappa系数为0.94。水稻的制图精度为98.41%，用户精度为97.64%；玉米的制图精度为95.38%，用户精度为97.89%。其中总体精度和Kappa系数较最大似然法分类结果分别提高了5.28%和0.08。所提研究方法可为其他地区农作物高分数据作物类型制图提供借鉴。

Extraction Method of Crop Planted Area Based on GF-1 WFV Image

Obtaining planted area of crop has important significance for guaranteeing nation grain safety. The Farm NO.597, located in Baoqing County, Shuangyashan City, Heilongjiang Province was selected as an example to extract rice and maize planted area by taking WFV (Wide field view) sensor carried on GF-1 satellite with the spatial resolution of 16 m as data source, using the image produced on October 30, 2014, and calculating different characteristic bands. Firstly, the multi-characteristic data set was established based on the NDVI (Normalized difference vegetation index) calculated from the source image and the first three principal components analyzed by PCA (Principal component transform). Then, using the difference between different surface features in each characteristic band, the decision tree was built based on CART (Classification and regression trees) to classify rice and maize. The results showed that the overall classification accuracy was 96.15% and the Kappa coefficient was 0.94. Producer accuracy of rice was 98.41% and user accuracy was 97.64%. Producer accuracy of maize was 95.38% and user accuracy was 97.89%. This method provides the reference value for crop type mapping using GF-1 data in other agricultural areas.