【摘要】 植被是湿地生态系统健康状况的"晴雨表",明晰湿地中植被的时空分布,是湿地修复与重建、保护与合理利用的前提和基础。以洪河国家级保护区为研究区,利用全极化C-band Radarsat-2和L-band PALSAR数据,根据极化合成孔径雷达(synthetic aperture radar,SAR)目标分解理论,提取了该保护区不同波长的极化分解参数和特征参量,整合为多源极化SAR数据集,利用多尺度迭代分割算法和Random Forest机器学习算法,构建了研究区中植被的遥感识别模型,实现了对研究区中植被的高精度分类,并对比分析了不同频率SAR数据集在植被识别精度上的差异。研究结果表明,利用整合PALSAR和Radarsat-2极化数据集,获取的植被遥感分类结果的总体分类精度为86.77%,比利用PALSAR极化数据集的分类结果精度提高了15%,但是其与利用Radarsat-2极化数据集的分类结果精度差异不显著;浅水草本沼泽的生产精度达到了90.91%,深水草本沼泽的用户精度为90.63%;C-band PALSAR数据比L-band PALSAR数据更适用于高精度识别洪河国家级自然保护区中的植被。更多还原

【Abstract】 Vegetation is a"barometer"of health status of wetland ecosystem. Exploration of the spatial-temporal distribution of vegetation is the premise and foundation of wetland ecological restoration and reconstruction, wetland resource protection and rational utilization. This paper extracted different wavelength polarimetric decomposition parameters of Honghe National Nature Reserve based on polarimetric synthetic aperture radar(SAR) target decomposition theory, and integrated C-band Radarsat-2 with L-band PALSAR for multisource polarimetric parameters dataset. Multi-scale iterative segmentation algorithm and Random Forest machine-learning algorithm were utilized to classify land cover types in the reserve. This research further analyzed differences between multi-frequency SAR datasets in the identification accuracy of vegetation types in the reserve. The results showed that the integrating PALSAR and radarsat-2 polarization datasets achieved86.77% overall classification accuracy, which was 15% higher than that using PALSAR polarization images.However, there was no significant difference between classification result using radarsat-2 polarization images and that using combined SAR images. The production precision of shallow-water marshes achieved 90.91%.The user precision of deep-water marshes obtained 90.63%. C-band polarimetric SAR was more suitable than L-band polarimetric SAR for vegetation classification in Honghe National Nature Reserve.更多还原