In Synthetic Aperture Radar (SAR) Interferometry, accuracy and computational complexity issues arise in all the processing chain leading to the evaluation of the interferograms, namely, both in the formation of the individual images of the pair and in their possible coregistration [1]. In particular and concerning coregistration, resampling one complex SAR image involves interpolation kernels whose kind and length must be chosen according to a trade-off between accuracy and computational efficiency. In some interferometric SAR (InSAR) applications, as the estimation of ground-surface deformations [2], high order and long interpolation kernels are generally required. As a consequence, during the recent years, some interest has been addressed to the selection of convenient interpolation kernels for SAR image coregistration [3,4], the main interest being to determining the kernel producing the best interpolation accuracy for a given kernel length. At the same time, efficient implementations of the coregistration step in terms of memory requirements and flop-counts have been also proposed [5]. Recently, a backprojection algorithm for the fast and accurate focusing of SAR images has been developed [6]. Differently from other approaches [7], it is based on the use of a Non-Uniform FFT (NUFFT) routine [8] and on a CUDA implementation on a Graphics Processing Unit (GPU) in CUDA language [9]. The NUFFT-based backprojection has proven drastically more accurate than other approaches based on the use of FFTs and interpolators typically used in SAR, without a significant increase in the computational cost. A serious advantage of the backprojection algorithm is the possibility of achieving SAR focused images on arbitrary lattices. In this way, the images forming the interferometric pair can be directly calculated on a common computational grid, without the need of an explicit coregistration step. Alternatively, the coregistration can be achieved, following the use of standard SAR focusing algorithm, by NUFFTs. In this paper, we show how, by the algorithm developed in [6], it is possible to quickly achieving interferometric pairs with high degrees of coherence, the typical parameter used to evaluate the interferometric phase quality. Results on airborne data provided by the Air Force Research Laboratory are shown.
Image Formation and Coregistration via Non-uniform FFTs in SAR Interferometry / Capozzoli, Amedeo; Curcio, Claudio; Liseno, Angelo. - (2012), pp. 761-761. (Intervento presentato al convegno Progress in Electromagnetic Research Symposium 2012 tenutosi a Moscow, Russian Federation nel Aug. 19-23, 2012).
Image Formation and Coregistration via Non-uniform FFTs in SAR Interferometry
CAPOZZOLI, AMEDEO;CURCIO, CLAUDIO;LISENO, ANGELO
2012
Abstract
In Synthetic Aperture Radar (SAR) Interferometry, accuracy and computational complexity issues arise in all the processing chain leading to the evaluation of the interferograms, namely, both in the formation of the individual images of the pair and in their possible coregistration [1]. In particular and concerning coregistration, resampling one complex SAR image involves interpolation kernels whose kind and length must be chosen according to a trade-off between accuracy and computational efficiency. In some interferometric SAR (InSAR) applications, as the estimation of ground-surface deformations [2], high order and long interpolation kernels are generally required. As a consequence, during the recent years, some interest has been addressed to the selection of convenient interpolation kernels for SAR image coregistration [3,4], the main interest being to determining the kernel producing the best interpolation accuracy for a given kernel length. At the same time, efficient implementations of the coregistration step in terms of memory requirements and flop-counts have been also proposed [5]. Recently, a backprojection algorithm for the fast and accurate focusing of SAR images has been developed [6]. Differently from other approaches [7], it is based on the use of a Non-Uniform FFT (NUFFT) routine [8] and on a CUDA implementation on a Graphics Processing Unit (GPU) in CUDA language [9]. The NUFFT-based backprojection has proven drastically more accurate than other approaches based on the use of FFTs and interpolators typically used in SAR, without a significant increase in the computational cost. A serious advantage of the backprojection algorithm is the possibility of achieving SAR focused images on arbitrary lattices. In this way, the images forming the interferometric pair can be directly calculated on a common computational grid, without the need of an explicit coregistration step. Alternatively, the coregistration can be achieved, following the use of standard SAR focusing algorithm, by NUFFTs. In this paper, we show how, by the algorithm developed in [6], it is possible to quickly achieving interferometric pairs with high degrees of coherence, the typical parameter used to evaluate the interferometric phase quality. Results on airborne data provided by the Air Force Research Laboratory are shown.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.