基于InSAR和深度学习的珠海市沉降区域识别与预测OA

This paper takes Zhuhai City as the research area,and integrates time-series InSAR technology with Deep Learning algorithms to conduct large-scale,high-precision research on surface subsidence identification and prediction.A dataset comprising 58 Sentinel-1A SAR images acquired from January 2022 to December 2023 is processed using a combined PS-InSAR and SBAS-InSAR approach to derive temporally dense surface deformation time series.Three distinct subsidence areas are identified,which are Baijiao Town in Doumen District,the under-construction segment of Xianghai Expressway in Jinwan District,and Pingsha Town,exhibiting average subsidence rates of-36.00 mm/yr,-37.20 mm/yr,and-37.94 mm/yr,respectively.The monitoring data indicate that Zhuhai City maintains overall geomechanical stability,with localized subsidence associated with infrastructure development,groundwater activities,and specific soil conditions.The LSTM-based Deep Learning model is innovatively developed using PS-InSAR-derived subsidence time series as input.The model is trained and validated in representative areas,achieving a relative error between predicted and actual values ranging from 0.52%to 5.83%.Both the MAE and RMSE on the test set remain below 4.7 mm.The model effectively captures subsidence trends and delivers high-precision predictions,providing a scientific basis and technical support for geological disaster prevention and planning in Zhuhai and other coastal cities.