Unsupervised Change Detection in Polarimetric SAR Data With the Hotelling-Lawley Trace Statistic and Minimum-Error Thresholding

Citation

Ghanbari M & Akbari V (2018) Unsupervised Change Detection in Polarimetric SAR Data With the Hotelling-Lawley Trace Statistic and Minimum-Error Thresholding. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 11 (12), pp. 4551-4562. https://doi.org/10.1109/jstars.2018.2882412

Abstract
Increased discrimination capability provided by polarimetric synthetic aperture radar (PolSAR) sensors compared to single and dual polarization synthetic aperture radar (SAR) sensors can improve land use monitoring and change detection. This necessitates reliable change detection methods in multitemporal PolSAR datasets. This paper proposes an unsupervised change detection algorithm for multilook PolSAR data. In the first step of the method, the Hotelling-Lawley trace (HLT) statistic is applied to measure the similarity of two multilook covariance matrices. As a result of this step, a scalar test statistic image is generated. Then, in the second step, a generalized Kittler and Illingworth (K&I) minimum-error thresholding algorithm is developed to perform on the test statistic image and discriminate between changed and unchanged areas. The K&I thresholding algorithm is based on the generalized Gamma distribution for statistical modeling of change and no-change classes. The proposed methodology is tested on a simulated PolSAR data and two C-band fully PolSAR datasets acquired by the uninhabited aerial vehicle SAR and RADARSAT-2 SAR satellites. The experiments show that the proposed algorithm accurately discriminates between change and no-change areas providing detection results with noticeably lower error rates and higher detection accuracy values compared to those of a CFAR-type thresholding of the HLT statistic. Also, the performance of the HLT statistic compared to the other statistics applied on the multilook polarimetric SAR data is assessed. Goodness-of-fit test results prove that the estimated generalized Gamma class conditional models adequately fit the corresponding change and no-change classes.

Journal
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing: Volume 11, Issue 12