Unsupervised change detection from multitemporal multichannel SAR images based on stationary wavelet transform

Synthetic aperture radars (SARs) have demonstrated their high efficiency in environment monitoring and disaster management, due to the short revisit time provided by them, their ability to operate in day and/or night and their independence of weather conditions. One of their applications that receives a lot of attention in the last few years is the change detection of the observed earth surface by exploiting the multitemporal multichannel (multipolarization and/or multifrequency) SAR images. In this paper, we propose an unsupervised process for the change detection from multitemporal multichannel SAR images that does not require any speckle filtering. This process is based on: i) generating a multiresolution set of each single-channel log ratio image using stationary wavelet transform (SWT); ii) applying the T-point algorithm for all the images of the multiresolution sets; iii) fusing all the channels using linear discriminant analysis (LDA) for each scale; and iv) fusing the obtained images to generate the change map. The proposed process was experimentally validated using semisimulated and real dual polarimetric images acquired by RADARSAT-2 satellite.