Multi-channel InSAR Elevation Reconstruction Research Based On Deep Learning

Interferometric Synthetic Aperture Radar(InSAR)is a remote sensing mapping technology combined with synthetic aperture radar and radio astronomical interferometry,which can accurately reconstruct digital elevation model(DEM)for ground surfaces.Phase unwrapping is a key step in InSAR data processing,which has a decisive impact on the accuracy of elevation reconstruction for ground surfaces.The multi-channel InSAR elevation reconstruction technique is based on the single-channel InSAR phase unwrapping technique,and it can reduce the impact of terrain complexity on elevation reconstruction accuracy for ground surfaces through joint processing multiple interferograms acquired.Deep learning has a strong function fitting ability,making it a powerful advantage in the field of multi-channel InSAR height reconstruction.Multi-channel InSAR elevation reconstruction technique is studied in this paper,and its main work as follows:(1)Related principles are introduced for InSAR and multi-channel InSAR,and then the multi-channel elevation reconstruction algorithms are investigated,and the experiments for elevation reconstruction are carried out to compare the performances of various algorithms.(2)A multi-channel InSAR elevation reconstruction method based on a multi-channel feature connected and fusion net,named as MCFCF Net,is proposed.The MCFCF Net is a neural network that maintain symmetrical structure and can directly establish the mapping relationship between multi-channel interferograms and elevation map for observed scenes,and use associated information between multi-channel interferograms to accurately restore elevation.The proposed method has obtained fine reconstruction results in different types of elevation reconstruction experiments.(3)A multi-channel InSAR elevation reconstruction method based on a dual-stream network which consists of elevation reconstruction stream and boundary detection stream,is proposed to obtain elevation information of observed scenes.The boundary detection stream can detect the boundary information of different terrains and merge it into the elevation reconstruction stream to improve the accuracy of elevation estimates for observed scenes.(4)An InSAR phase filtering method based on a dual-stream network which consists of phase filtering stream and fringe detection stream,is built to suppress phase noise present in interferograms.On this basis,a multi-channel InSAR elevation reconstruction method inserted with phase filtering is proposed,and the effectiveness of the proposed method is verified in the experiments of elevation reconstruction for different observed scenes.