Abstract: The reasonable development of tea plantations is of great significance for the local economy and ecological environment protection. Obtaining accurate spatial distribution of tea plantations is an important factor in tea plantations management and land use. This paper takes Lianhua Town, Tong'an District, Xiamen City, Fujian Province as the research area, and takes Sentinel-2 remote sensing image as the data source to analyze the difference of spectral characteristics of multi-temporal Sentinel-2 spectrum in different ground objects. The vegetation indices with sensitive spectrum and texture features were used as the extraction characteristics of tea plantations. The distribution range of tea plantations in the research area was extracted using the random forest classification method. The results showed that the random forest method combined with Sentinel-2 spectral and texture information can achieve good results in many periods. Among them, the overall accuracy and Kappa coefficient in October were the highest, reaching 95.53% and 0.9088, respectively; the classification accuracy in April and December was relatively close, with overall accuracies of 94.28% and 94.41%, and Kappa coefficients of 0.8889 and 0.8881, respectively; the classification accuracy in July was the lowest, with an overall accuracy of 91.05% and a Kappa coefficient of 0.8254. The classification results meet the research needs.

Key words: remote sensing, tea plantations, Sentinel-2, random forest

摘要： 以福建省厦门市同安区莲花镇为研究区，以Sentinel-2 遥感影像为数据源，分析多时相的Sentinel-2光谱在不同地物上的光谱特征差异，选择敏感光谱构建的植被指数和纹理特征作为茶园提取特征，使用随机森林分类方法提取研究区茶园的分布范围。结果表明，随机森林方法结合Sentinel-2光谱和纹理信息，在多个时期都能达到较好的提取效果。其中，在10月的总体精度和Kappa系数最高，分别为95.53%和0.9088；4月和12月的分类精度比较接近，总体精度分别为94.28%和94.41%，Kappa系数分别为0.8889和0.8881；7月的分类精度最低，总体精度为91.05%，Kappa系数为0.8254。分类结果满足茶园管理研究需要。

关键词: 遥感, 茶园, Sentinel-2, 随机森林