Change Detection in Multi-Temporal SAR Images Based on Slow Feature Analysis Combined With Improving Image Fusion Strategy
Weisong Li,Xiayang Xiao,Penghao Xiao,Haipeng Wang,Feng Xu
DOI: https://doi.org/10.1109/jstars.2022.3166234
IF: 4.715
2022-01-01
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Abstract:Change detection in multitemporal synthetic aperture radar (SAR) images has been an important research content in the field of remote sensing for a long time. In this article, based on the slow feature analysis (SFA) theory and the nonsubsampled contourlet transform (NSCT) algorithm, we propose a novel unsupervised change detection method called NSCT nonlocal means (NSCT-NLM). The powerful extraction to the changed information of SFA and the superior information fusion of NSCT are jointly adopted in this method. The main framework consists of the following parts. First, SFA and the log-ratio operator are used to generate difference images (DIs) independently. Then, the NSCT is used to fuse two DIs into a new higher quality DI. The newly fused DI combines the complementary information of the two kinds of original DI. Therefore, the contrast of the changed regions and unchanged regions is greatly enhanced, as well as the changed details are preserved more completely. Furthermore, an NLM filtering algorithm is employed to suppress the strong speckles in the fused DI. We use the fuzzy C-means algorithm to generate the final binary change map. The experiments are carried out on two public datasets and three real-world SAR datasets from different scenarios. The results demonstrate that the proposed method has higher detection accuracy by comparing with the reference methods.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geography, physical