Time-referenced wavelet spatio-temporal change detection on multi-temporal SAR images

Abstract

We discuss here an unsupervised method to detect spatio-temporal changes on multitemporal SAR images. We name it TR-WECS (Time-Referenced Wavelet Energies Correlation Screening). We have a series of SAR images taken over time on the same region. We establish the first image as our time reference against which changes will be studied. We employ wavelet smoothing in each image, so that detected changes are not speckle-related. We then analyze temporal changes on discontiguous blocks of various sizes through wavelet energy apportionment, and ultra-high dimensional correlation screening. We illlustrate TR-WECS on a time series of Sentinel-1 images on the border of Brazil and French Guiana. The proposal's performance is favorably compared to traditional change detection methods.