哨兵一号协同吉林一号影像的树种识别研究

Study on Tree Species Identification by Combining Sentinel-1 and JL101A Images

针对C波段合成孔径雷达影像协同光学影像进行森林分类时失相干现象较为严重的问题,提出将冬季相近时相合成孔径雷达的相干系数影像进行均值处理的方法,有效抑制了失相干现象,得到与森林树种相关的相干系数影像。本文分别利用Sentinel-1后向散射强度、相干系数和吉林一号光学A星影像,对长春市净月潭森林公园的树种进行分类研究。结果显示:在仅用吉林一号光学A星数据进行分类时,总体分类精度为82.3%,Kappa系数为0.79;在使用吉林一号数据和哨兵一号强度数据进行分类时,总体分类精度为85.2%,Kappa系数为0.825;在使用吉林一号数据、哨兵一号强度数据和相干性数据进行分类时,总体分类精度为87.8%,Kappa系数为0.855;在使用吉林一号数据、哨兵一号强度数据和相干性数据进行分类以后,相较于仅使用吉林一号数据,落叶松用户精度由原来的59%提升到了72%,表明光学影像结合C波段合成孔径雷达影像可提高森林树种分类精度。

Aiming at the serious problem of de-coherence in the forest classification of the C-band synthetic aperture radar image in conjunction with the optical image, a method of averaging coherence coefficient images in multiple periods in winter obtained by coherent imaging of similar time-phase radar images is proposed to effectively suppress the de-coherence phenomenon to obtain coherence coefficient images related to forest tree species. This paper uses the Sentinel-1 backscattering intensity, coherence coefficient and JL101 A image to classify the tree species of the moon lake national forest park in Changchun. The results show that when using only JL101 A for classification, the overall classification accuracy is 82.3%, Kappa coefficient is 0.79;when using JL101 A data and Sentinel-1 intensity data for classification, the overall classification accuracy is 85.2%, and the Kappa coefficient is 0.825;when using JL101 A data, Sentinel-1 intensity data and coherence data for classification, the overall classification accuracy is 87.8%, and the Kappa coefficient is 0.855. After using JL101 A data, Sentinel-1 intensity data and coherence data to classify, compared with using only JL101 A data, the precision of user accuracy is increased from 59% to 72%. It shows that the optical image combined with the effective coherence coefficient and backscattering intensity of C band synthetic aperture radar image can improve the classification accuracy of forest tree species.