基于时间序列Sentinel-1A数据的玉米种植面积监测研究

Monitoring of maize planting area based on time-series Sentinel-1A SAR data

玉米作为中国三大作物之一,监测其种植面积对及时了解其种植时空分布、保障粮食安全具有重要作用。本文以河北省涿州市为研究区,利用2016年多时相Sentine-1A SAR(Synthetic Aperture Radar,合成孔径雷达)影像,对玉米种植区域进行提取。在对研究区地物散射特性分析的基础上,分析了微波后向散射特性随不同生育期玉米植株结构发育的变化情况,选择合适时相和极化组合的后向散射系数,运用支持向量机(Support Vector Machine,SVM)算法提取了玉米种植范围和面积信息,并对不同后向散射系数(标准后向散射系数(Sigma-naught,σ0)和归一化后向散射系数(Gamma-naught,γ0))用于研究区作物种植区提取的结果进行了比较。结果表明,采用时间序列(4月19日,5月30日,6月11日,7月17日)雷达图像得到的监督分类结果具有较高的分类精度和kappa系数,总体精度达92.96%,Kappa系数为0.91。因此,采用4—7月(春玉米播种至吐丝时期、夏玉米播种至拔节时期)的时间序列SAR数据能有效获取不同种植模式下的玉米信息,而增加8、9月的数据对玉米识别精度的影响不大。总体来说,采用多时相双极化的σ0数据与相同时相组合的γ0数据对玉米种植范围提取基本相同,但使用γ0数据的林地识别精度比σ0数据提高了3%。研究结果可为多极化SAR数据的玉米识别和面积监测提供参考案例。

Maize is one of the three typical crops in China. Monitoring the distribution of maize is of vital importance in understanding the scope of corn cultivation and ensuring food security. Multi- temporal ESA Sentinel- 1A C-band Synthetic Aperture Radar （SAR） VV and VH polarization data at a 20 m spatial resolution were used to identify maize identity for Zhuozhou City in Hebei province, China. Based on the analysis of the scattering characteristics for different crops, as well as the backscattering coefficients variability of different crop structures at different growth and development stages of the study area, dual-polarized backscatter coefficients from appropriate phases were selected. To assess the performance of different backscattering coefficients forms in classification, both σ0 and γ0 data were used. Compared with the sampling data from the field campaign, the results showed that a high recognition accuracy of maize can be obtained using support vector machine （SVM） method if time series data are used. Classification result obtained from time series radar images has a higher accuracy and kappa coefficient than those from a single image. For example, results gotten from long time series o-0 data have the ideal performance with overall accuracy of 92.96% and a Kappa coefficient of 0.91. Data from sowing to silking of spring maize and sowing to jointing of summer maize （April 19, May 30, June 11, and July 17） can effectively obtain maize information under different cropping patterns. The increase data of August and September would have little impact on the accuracy of maize identification. The differences between σ0 and γ0 data are relatively moderate. However, the classification accuracy obtained with long time series γ0 data can be improved by 3% than that with σ0 data for the forest area.