Assessment of Nonidentical Flight Passes on Wavelength-Resolution SAR Change Detection Based on Tensor Robust PCA

Change detection methods are, usually, restricted to SAR image pairs or stacks from identical flight geometries. This selection is mainly explained by the fact that the specular reflection of each object on the ground, such as vehicles, power lines, and buildings, depends on the incident angle and, therefore the heading angle of the platform in a pass. Thus, non-identical acquisitions can significantly increase the number of false alarms for change detection methods. This paper assesses non-identical flight passes on change detection based on tensor robust principle component analysis (TRPCA). We have considered the wavelength-resolution SAR images from the CARABAS-II data set for this assessment. These SAR images are well known for their stability and for being unaffected by speckle noise, thanks to the system resolution in the order of the radar signal wavelength. The experiments in four cases provided in the paper, two cases using images with identical passes and two cases using non-identical passes, have shown that TRPCA can perform very well change detection even in the scenarios where a surveillance image was acquired with a flight pass different from the ones used for the reference images. In addition, the results have shown that the most significant contribution of false alarms comes from elongated structures that can be sensitive to a flight pass but insensitive to others. Such elongated structures can be removed through post-processing techniques.