Abstract: With the increasing expansion of greenhouses and the utilization of plastic films in agricultural production, rapid and accurate monitoring of their spatial distribution has become an important foundation for agricultural management. Based on the Google Earth Engine (GEE) cloud platform, we integrated Sentinel-2 remote sensing data with a Random Forest algorithm. Through feature selection, spectral, index, and optimal texture features (G+Z+W2) were identified as the optimal feature combination to achieve rapid and accurate extraction of greenhouses and plastic films in Liaoning Province. The results showed that the proposed algorithm effectively interpreted greenhouses and plastic films. The overall classification accuracies in 2019 and 2024 were 91.3% and 89.21%, with Kappa coefficients of 0.84 and 0.81, respectively. From 2019 to 2024, greenhouse area increased from 1286.61 to 2005.63 km², and plastic film area increased from 407.06 to 713.38 km². The aggregation index of greenhouses increased from 70.99 to 72.37, and the mean patch area increased from 0.88 to 1.11 hm². The aggregation index of plastic films increased from 58.06 to 64.8, and the mean patch area increased from 0.48 to 0.68 hm², indicating a transition from dispersed to clustered distribution and from small-scale to regionally concentrated. In terms of spatial distribution, greenhouses were mainly distributed in Gaizhou and Beizhen. Influenced by temperature and spring precipitation, plastic films were mainly concentrated in Jianping and Jinzhou in northwestern Liaoning. This study provides an efficient and reliable technical methodological framework for regional-scale monitoring of agricultural plastic coverage and offers a data foundation for agricultural management.

Key words: greenhouse extraction, plastic film identification, Google Earth Engine, random forest, Sentinel-2