A change detection statistic for repeat pass interferometric SAR

Physics-Based Predictions for Coherent Change Detection Using X-Band Synthetic Aperture Radar

“…In practice, this lower level is modelled as an equivalent coherency noise, and observationally this noise level is equivalent to approximately 7.2 dB below the HH backscattering coefficient [1]. Such a level of effective noise leads to a reduced coherency of γ = γ scene γ snr eff = γ scene 1 1 + σ n /σ HH = 0.84 (24) for the case of perfect scene coherence. This level of effective noise has been modelled in the calculations by adding It is well known that the sample coherency evaluated over a finite window is a biased estimator of the true underlying scene coherency [19].…”

“…ROCs calculated in this way are displayed in Figure 6. Figure 6 also shows theoretical ROC curves for a loglikelihood change statistic [1,24,25] that may be evaluated for the case when the full 2 by 2 sample covariance matrix describing the interferometric image pair is available for discriminating between the changed and unchanged scenarios. The log-likelihood change statistic is based on the assumption that the N-pixel region under test in the repeat-pass image pair is described by a complex, circular, jointly Gaussian random process.…”

“…The log-likelihood statistic (LLS) ROC curves of Figure 6 thus show the theoretical performance that may be obtained using a 3 by 3 spatial estimation pixel window applied to the simulated changed and unchanged scenes. The improvements in detection performance may be attributed to the use of the complete 2 by 2 covariance matrix describing the repeat-pass image pair to discriminate between the changed and unchanged hypotheses [26].…”

“…The log-likelihood change statistic is based on the assumption that the N-pixel region under test in the repeat-pass image pair is described by a complex, circular, jointly Gaussian random process. It is obtained by forming the likelihood ratio defined as the ratio of the probability that the N pixel pairs in the local area under test are realisations of the unchanged hypothesis to the probability that the N pixel pairs are realisations of the changed hypothesis [24].…”

Urban change detection using multitemporal ERS-1/2 InSAR data

“…Particularly the interferometric correlation, as a measure for the variance of the interferometric phase, provides complementary information about the properties of a 'scene' to the intensity of backscattering [4]. It can reveal reliable information about the land-cover changes that may not be evident in the intensity data alone [1], [5], [6].…”