基于NDVI物候特征的华南地区冬种马铃薯遥感提取方法

Remote Sensing Method Based on Multi-temporal NDVI Phenological Characters for Winter Potato Planting Area in South China

马铃薯是华南地区的特色冬种农作物,其地块的"早稻―晚稻―冬种马铃薯"三季种植模式具有特有的植被指数时间序列曲线特征。利用这一特征,提出一种基于NDVI时间序列数据和SAM的冬种马铃薯种植面积提取方法。以广东省惠州市稔平半岛为研究区,冬种马铃薯面积为研究对象,采用2011年HJ-1 A/B CCD遥感数据为主要数据,计算每一景影像的NDVI后以时间为坐标轴排列成NDVI时间序列数据集,在此提取冬种马铃薯种植区的NDVI时间序列参考曲线,使用光谱角度匹配(SAM)方法,计算每个像元的NDVI时间序列曲线与NDVI时间序列参考曲线的光谱夹角值,根据Rule图像的统计参数确定夹角阈值,达到快速有效地在遥感影像上提取冬种马铃薯对应像元的目的。结果表明:研究区总体提取精度为82.70%,重点种植区域提取精度为93.75%,可见基于NDVI物候特征的SAM方法能够有效提取研究区冬种马铃薯的种植面积。

Vegetation index of each pixel in remote sensing image reflects the growth state of vegetation covered by pixels. If the vegetation index images obtained from all remote-sensing images in one year in the same area are lined up, the dynamic change of vegetation index time series can mirror the phenological rules in this area. Potato is a special kind of winter-planted crop in South China, and relevant parcels show peculiar curve characters of vegetation index time series in planting pattern-＂early season rice-late season rice-winter-planted potato＂. Taking Renping Peninsula in Huizhou City, Guangdong Province, as research area and the planting area of winter potato as research object, the paper makes use of those characters and employs HJ image data in 2011 as the primary data, with totaling 14 scenes＇ HJ-1 A/B CCD remote sensing image data being used. After calculating the NDVI （Normalized Difference Vegetation Index） of images in each scene, NDVI time serials data set is obtained. Corresponding NDVI time serials curve of 26 training sample points is achieved for averaging, thus getting NDVI time serials character reference curve. Moreover, SAM （Spectral Angle Mapping） is used to calculate the value of spectral angle between NDVI time series curve of each pixel in the NDVI time serial data set and NDVI time series character reference curve, obtaining the image of spectral angle value. The pixel value in this image is the value of spectral angle between NDVI time serials curve of each pixel in the NDVI time serial data set and character reference curve. The 2-time standard deviation, 2.5-time standard deviation and 3-time standard deviation of the image of spectral angle value are taken as the threshold value. The spectral angle value of each pixel in the image of spectral angle value and the threshold value are compared, and if the spectral angle value is less than the threshold value, this pixel is classified as a target object, so as to separate the target pixel from pixels covered by other types of land on the remote sensing images, for the purpose of rapidly and effectively extracting the planting area of winter potato. According to the comparative analysis, it is found that the result extracted by taking 2.5-time standard deviation as the threshold value fits best the actual planting conditions, thereby serving as the final extraction result. Research findings show that the overall extraction accuracy is 82.70% in the research area and 93.75% in key planting areas. This method, as one of the typical applications of homemade optical satellite to agriculture in South China, can extract the winter potato area effectively and will lay a firm foundation for accurately and rapidly monitoring other agricultural information （such as winter-fallowed cultivation） in South China where there is much cloud, rain and broken land.