Page 85 - 《武汉大学学报（信息科学版）》2025年第9期

第 50 卷第 9 期 郭浩然等：基于样本迁移的无监督植被遥感制图方法 1813

sults: The experiment select Arhorqin Banner as the key study area, and complete the regional multi-tem⁃
poral natural vegetation classification and mapping from 2005 to 2022, with the overall accuracy of the clas⁃
sification better than 88%, and the Kappa coefficient greater than 0.80. The natural vegetation in the study
area is dominated by grasslands, forests and scrubs are mainly distributed in the northwestern part of the
study area, with temperate tufted grassland dominating the grasslands and the temperate tufted grassland
dominated by the extensive needlegrass grassland. From 2005 to 2022, natural vegetation shrinkage in⁃
creases from north to south, while temperate deciduous broad-leaved forests show little change. The most
pronounced changes occur in the southern part of the study area, where temperate graminoid grasslands and
temperate deciduous scrub decline steadily, and non-natural vegetation expand year by year. Further analy⁃
sis of the situation shows that the southern part of the study area is dominated by sandy vegetation, which
has been seriously degraded since 2000. Still, the study area has started to implement the artificial forage in⁃
dustry since 2010, so the natural vegetation area has decreased, and the unnatural vegetation area has in⁃
creased. The unnatural vegetation area is mostly small round patches, which are more concentrated in the
southern part of the study area. Conclusions: The overall accuracy of the classification meets the needs of
long-time series vegetation classification mapping, and the mapping effect is more stable. Therefore, the
unsupervised sample migration method based on historical large-label vegetation classification maps can
make full use of the existing vegetation classification products, and provide a more convenient way for vege⁃
tation classification mapping update. By using the geometric features of the historical vegetation classifica⁃
tion products and combining with the multi-source remote sensing data of the same or similar time-phase,
we construct the feature attribute set of the patch-by-patch, and hierarchically optimize samples of the
large-label data of vegetation classification from local to global, to form the optimized and migratable vege⁃
tation classification maps. Optimization of samples is carried out to form a migratable training set of opti⁃
mized vegetation classification samples. The optimized vegetation classification samples training set is then
combined with land use data to carry out multi-temporal natural vegetation classification and mapping, and
the results of long-time series natural vegetation classification changes are obtained. This study provides a
fast, convenient, lightweight, and reliable mapping method for vegetation remote sensing classification.
Key words： long-time series remote sensing； sample optimization； sample migration； unsupervised classi⁃
fication； vegetation mapping

地球表面上生长的所有植物群落统称为植 结合植被的分布特征对生态系统空间格局进行
被，作为反映生态环境变化的重要依据，植被对 相关分析；文献［8］在土地利用/覆被和植被遥感
地 球 生 态 系 统 和 人 类 社 会 有 着 极 其 重 要 的 作 监测等方面开创应用先河；文献［9］利用归一化
［1］
用 。植被分类是通过植物物种的生长环境、物 植被指数（normalized difference vegetation index，
种组成和生长状况等特征，建立一套统一的分类 NDVI）对 1982—2015 年哈萨克斯坦的植被信息
准则，准确的植被分类为合理利用和有效保护植 （农 田 、草 地 和 灌 丛）进 行 了 提 取 及 变 化 特 征 分
被提供科学的依据 。因此，获取植被类别的分 析；文献［10］结合实地采样数据，从遥感图像中
［2］
布和生长状况并进行植被分类制图，对开展生态 提取不同类别草地的光谱、纹理和空间特征，根
环境保护工作有着极其重要的意义。自然界中， 据一定的规则将图像中的每个像素划分为不同
植被的类型复杂多样，传统植被分类制图方法主 的类别；文献［11-13］使用长时序反映植被生长状
要依靠人工。随着新型遥感数据和分类算法的 况 的 植 被 指 数 ，如 归 一 化 燃 烧 指 数（normalized
发展，植被遥感制图的精度和效率不断提高，为 burn ratio，NBR）和反映图像同质性的纹理特征，
植被分类提供一种新方法，并因为其快速和高效 结合原始多光谱图像及地形特征、气象特征等数
被广泛应用 ［3-6］ 。 据进行植被分类，成功改进了植被分类精度。
监督分类因其可以生成准确的分类结果，是 目前的植被遥感分类制图方法大多依赖于
目前植被分类制图中应用最为广泛的方法。文 单一时相的实地数据，但实地采样往往需要耗费
献［7］制作土地覆盖分类和植被指数分布图，并 大量的人力物力，因此仅依靠实地样点数据实现