Abstract: Using satellite remote sensing technology coupled with deep learning algorithms could characterize open-pit coal mining area dynamically and efficiently. This study focused on typical open-pit coal mining areas from China and other major coal-resource countries as research objects. Based on GF-2 multi-spectral remote sensing images， data sets and labels are produced to construct faster-regions with convolutional neural networks(CNN) features target recognition model. Low-resolution semantic layer and high-resolution texture information of mining area and background area were fully mined by incorporating the feature Pyramid network， which could improve faster-regions with CNN features model by optimizing parameters. The results showed that the average detection accuracy was improved to 98.48%， and the overall recognition accuracy reached 96.7%. The improved faster-regions with CNN features model efficiently increases the identification accuracy of multi-scale and multi-type open-pit mining targets in complex backgrounds. These findings can provide a scientific and accurate technical way for global energy cooperation， environmental protection， and mineral resources utilization in China.

Key words: domestic high-resolution imagery; open-pit coal mining areas; object detection; feature pyramids