Leveraging power of deep learning for fast and efficient elite pixel selection in time series SAR interferometry

Abstract

This study proposes a new convolutional long short-term memory (ConvLSTM) based architecture for selection of elite pixels (i.e., less noisy) in time series interferometric synthetic aperture radar (TS-InSAR). The model utilizes the spatial and temporal relation among neighboring pixels to identify both persistent and distributed scatterers. We trained the model on ~20,000 training images (interferograms), each of size 100 by 100 pixels, extracted from InSAR time series interferograms containing both artificial features (buildings and infrastructure) and objects of natural environment (vegetation, forests, barren or agricultural land, water bodies). Based on such categorization, we developed two models, tailormade to detect elite pixels in urban and coastal sites. Training labels were generated from elite pixel selection outputs generated from the wavelet-based InSAR (WabInSAR) software. We used 4 urban and 7 coastal sites for training and validation respectively, and the predicted elite pixel selection maps reveal that the proposed models efficiently learn from WabInSAR-generated labels, reaching a test accuracy of 94%. The models accurately discard pixels affected by geometric and temporal decorrelation while selecting pixels corresponding to urban objects and those with stable phase history unaffected by temporal and geometric decorrelation. The density of pixels in urban areas is comparable to and higher for coastal areas than WabInSAR outputs. With significantly reduced time computation (order of minutes) and improved density of elite pixels, the proposed models can efficiently process long InSAR time series stacks and generate deformation maps quickly, making the time series InSAR technique more suitable for varied (non-urban and urban) terrains and unaddressed land deformation applications.

Tasks

Reproductions