8.1. Omnibus Test Change Detection
Conradsen et al. (2016) present a change detection algorithm for time series of complex valued SAR data based on the complex Wishart distribution for the covariance matrices. \(S_{rt}\) denotes the complex scattering amplitude where \(r,t \in \{h,v\}\) are the receive and transmit polarization, respectively (horizontal or vertical). Reciprocity is assumed, i.e. \(S_{hv} = S_{vh}\). Then the backscatter at a single pixel is fully represented by the complex target vector
For multi-looked SAR data, backscatter values are averaged over \(n\) pixels (to reduce speckle) and the backscatter may be represented appropriately by the (variance-)covariance matrix, which for fully polarimetric SAR data is given by
where \({\langle \cdot \rangle}\) is the ensemble average, \(^*\) denotes complex conjugation, and \(^H\) is Hermitian conjugation. Often, only one polarization is transmitted (e.g. horizontal), giving rise to dual polarimetric SAR data. In this case the covariance matrix is
These covariance matrices follow a complex Wishart distribution as follows:
where \(p\) is the rank of \(\boldsymbol X_i = n {\langle \boldsymbol C_i \rangle}\), \(E[\boldsymbol X_i] = n \boldsymbol\Sigma_i\), and \(\boldsymbol\Sigma_i\) is the expected value of the covariance matrix.
In the first instance, the change detection problem then becomes a test of the null hypothesis \(H_0 : \boldsymbol\Sigma_1 = \boldsymbol\Sigma_2 = ... = \boldsymbol\Sigma_k\), i.e. whether the expected value of the backscatter remains constant. This test is a so-called omnibus test.
A test statistic for the omnibus test can be derived as:
where \(\boldsymbol X = \sum_{i=1}^k \boldsymbol X_i \sim W_C(p,nk,\boldsymbol\Sigma)\). The test statistic can be translated into a probability \(p(H_0)\). The hypothesis test is repeated iteratively over subsets of the time series in order to determine the actual time of change.
See Also:
References:
Conradsen, K., Nielsen, A. A., & Skriver, H. (2016). Determining the Points of Change in Time Series of Polarimetric SAR Data. IEEE Transactions on Geoscience and Remote Sensing, 54(5), 3007–3024.