Mapping agricultural plastic greenhouses using Google Earth images and deep learning

Abstract The worldwide use of agricultural plastic greenhouses (APGs) is crucial to provide sufficient food, including vegetables and fruits, for residents. However, the pollution problem created by plastic materials has also aroused widespread concern. Therefore, it is important to obtain the spatial distribution of APGs via different approaches, especially using remote sensing images. In this study, a deep learning method is adopted to map the distribution of APGs in Shouguang, Shandong Province, China, with high-resolution Google Earth images. The results suggest that the distribution of greenhouses can be accurately extracted with a mean intersection over union (mIOU) of 97.20%. The total area covered by APGs is 185.37 km 2 , and the total number of APGs is approximately 170,807. Both densely and sparsely distributed APGs can be extracted effectively. This research shows that the deep learning method can extract greenhouse information quickly and effectively from high-resolution images and can be used in agricultural pollution monitoring and agricultural development planning.