关键词: 花生, 种植信息, 谷歌地球引擎, 随机森林, 物候特征, 色素指数, 纹理特征

Abstract: Peanut is one of the important oil crops in China. Using remote sensing images to classify peanut cultivated land accurately and efficiently is of great significance for maintaining food security and guiding agricultural production. This research took Kangping County，Shenyang City as the research area，used Sentinel‐2 multi temporal remote sensing image as the data source based on Google Earth Engine（GEE）platform，constructed vegetation index features，texture features and phenological features，and used pixel‐based and object‐oriented random forest classifiers to achieve peanut planting information extraction in Kangping County，Shenyang City. The experimental results showed that the pigment index，texture feature and phenology feature improved the performance of the classifiers to a certain extent，among which the texture feature had the largest improvement for the object‐oriented random forest classifier.Compared with the use of spectrum+vegetation index+pigment index feature，the overall accuracy and Kappa coefficient after adding texture feature increased by 3.85 percentage points and 0.068 4 respectively，and the pixel‐based random forest classifier also rose by 2.66 percentage points and 0.035 6.In different feature combinations，the performance of pixel‐based random forest classifier was better than that of object‐oriented random forest classifier. The optimal overall accuracy of the two classifiers was 95.28% and 92.31% respectively，and the optimal Kappa coefficients were 0.908 6 and 0.860 8 respectively.

Key words: Peanut, Planting information, Google Earth Engine, Random forest, Phenological characteristics, Pigment index, Texture feature