Automatic identification of mining-induced subsidence using deep convolutional networks based on time-series InSAR data: a case study of Huodong mining area in Shanxi Province, China

Underground mining has the potential to trigger wide-area surface subsidence, and it is crucial to identify and investigate mining-induced subsidence to reduce potential damages. In recent years, the development of remote sensing and deep learning has brought a novel direction to the automatic identification of mining-induced subsidence. In this paper, we present a comprehensive solution for the automatic identification of mining-induced subsidence using deep convolutional networks based on time-series interferometric synthetic aperture radar (InSAR) data. First, we obtain a surface deformation rate map of the Huodong mining area based on SBAS-InSAR technology. Second, we automatically identify mining-induced subsidence areas from the InSAR deformation map based on a convolutional neural network (CNN) deep learning model. Finally, we summarize the potential risk of surface subsidence to roads in the Huodong mining area. The InSAR results show the detailed characteristics of surface deformation with a maximum subsidence rate of 112.1 mm/year, where the surface deformation is distinctly different between the urban, forest, and mining areas. The deep learning results show 202 identified mining-induced subsidence areas, including the subsidence areas located in the forest which were difficult to distinguish in the InSAR monitored deformation map. This paper demonstrates the potential and capability of deep learning models for more automatic and accurate identification of mining-induced subsidence and provides a comprehensive solution for improving the efficiency of interpreting surface deformation monitoring information.