2019/09/04 by Ashutosh Tiwari, Tiwari, Ashutosh, Avadh Bihari Narayan +3
Engineering · #Advanced SAR Imaging Techniques #FOS: Computer and information sciences #FOS: Electrical engineering #Geophysical Methods and Applications #Image and Video Processing (eess.IV) #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Synthetic Aperture Radar (SAR) Applications and Techniques #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1909.01868
openalex publication_date 2019/09/04 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
In multi-temporal SAR interferometry (MT-InSAR), persistent scatterer (PS)\npixels are used to estimate geophysical parameters, essentially deformation.\nConventionally, PS pixels are selected on the basis of the estimated noise\npresent in the spatially uncorrelated phase component along with look-angle\nerror in a temporal interferometric stack. In this study, two deep learning\narchitectures, namely convolutional neural network for interferometric semantic\nsegmentation (CNN-ISS) and convolutional long short term memory network for\ninterferometric semantic segmentation (CLSTM-ISS), based on learning spatial\nand spatio-temporal behaviour respectively, were proposed for selection of PS\npixels. These networks were trained to relate the interferometric phase history\nto its classification into phase stable (PS) and phase unstable (non-PS)\nmeasurement pixels using ~10,000 real world interferometric images of different\nstudy sites containing man-made objects, forests, vegetation, uncropped land,\nwater bodies, and areas affected by lengthening, foreshortening, layover and\nshadowing. The networks were trained using training labels obtained from the\nStanford method for Persistent Scatterer Interferometry (StaMPS) algorithm.\nHowever, pixel selection results, when compared to a combination of R-index and\na classified image of the test dataset, reveal that CLSTM-ISS estimates\nimproved the classification of PS and non-PS pixels compared to those of StaMPS\nand CNN-ISS. The predicted results show that CLSTM-ISS reached an accuracy of\n93.50%, higher than that of CNN-ISS (89.21%). CLSTM-ISS also improved the\ndensity of reliable PS pixels compared to StaMPS and CNN-ISS and outperformed\nStaMPS and other conventional MT-InSAR methods in terms of computational\nefficiency.\n