Study on remote sensing feature selection of green manure crop Astragalus sinicus based on multitemporal Sentinel-2 imagery

Astragalus sinicus is the main planting and utilization of green manure crop in southern China. Crop rotation planting of green manure could increase soil fertility, improve soil micro-ecological environment, and is conducive to the enhancement of farmland productivity, which is of great significance for protecting farmland ecological environment and promoting agricultural sustainable development. Compared with the traditional means of ground survey and statistics method, remote sensing technology has the characteristics of macroscopic and real-time, which could quickly obtain the distribution of regional scale Astragalus sinicus planting. How to establish distinguishable remote sensing features is the basis of its recognition and extraction. In this study, multi-temporal Sentinel 2 images covering the main growing period of the Astragalus sinicus from December 2019 to June 2020 were selected as data source. The variation of spectral characteristics of Astragalus sinicus were studied in the typical pilot area of green manure crop rotation in Liuhe District, Nanjing City, Jiangsu Province. On this basis, the multi-temporal vegetation index data set was constructed. The recursive feature elimination and random forest algorithm were used to analyze the influence of feature composition on recognition accuracy, and then the effective recognition feature set of Astragalus sinicus at different time phases was established. Through this study, we explored the recognition ability of multi-temporal Sentinel 2 images for Astragalus sinicus, and provided a theoretical basis for the study of regional-scale Astragalus sinicus remote sensing monitoring.