Agricultural Greenhouse Extraction based on Sentinel-2 images in Fujian Province

Plastic greenhouse is a widely used cultivation facility in agricultural production, which can effectively reduce disasters and increase production. Accurate acquisition of greenhouse coverage information provides scientific data support for optimizing agricultural production layout and promoting the development of facility agriculture. We obtained Sentinel-2 image data of Fujian Province in the spring, autumn, and winter from 2019 to 2020 through Google Earth Engine (GEE). We extracted a number of spectral, textural, and DEM features, and used the Gini index to conduct feature extraction. Finally, we used the Random Forest (RF) classifier to extract agricultural greenhouses in different seasons in Fujian Province. The results show that: (1) Combined with GEE and RF, the overall extraction accuracy of Fujian Province is 84%, implying that this method is suitable for large-scale greenhouse extraction. Among them, texture features have a significant impact on the classification of greenhouses. (2) Greenhouses are mainly distributed in Minnan Plain of Fujian Province, accounting for 45.88% of the total area of Fujian Province, but less in inland areas. (3) Comparing the area of greenhouses in different seasons, it was found that the largest area was covered by greenhouses in spring, followed by winter and autumn.