Change detection for ScanSAR images

ScanSAR images cover a vast area of 359 X 379 km and are taken for the same scene at intervals of 46 days. We found that, for a series of 12 consecutive images, the best recall of changes due to deforestation was obtained using the temporal standard deviation. However the change detection methods used for FBD images can also be applied to find changes between two ScanSAR images and these can be output as a colour map depicting levels of change. By stringing these together an animation of changes over time can be constructed. Furthermore we can either look at changes between time-adjacent images (which we call relative) or we can obtain the change between an image and the first in the series (which we call absolute). We performed this exercise for two overlapping ScanSAR scenes over central Sumatra which we called S1 (covering the East coast) and S2 (the more Westerly). The animations, which cover a period of about 18 months, are presented as powerpoint slideshows here:


S1 absolute animation

S1 relative animation

S2 absolute animation

S2 relative animation

The first animation reveals the changes due to deforestation (including the "axe") that we saw using FBD centre right of the image. But the area to the north shows that the effect of moisture and flooding is also very strong. The second, relative, animation should show transient changes more clearly but some of the deforestation regions are poorly represented. Moisture and flooding in the northerly region again is very apparent. I would surmise that these larger changes distort the detection parameters used make them less optimal for deforestation detection. This is the difficulty of detection over a large scale as opposed to localised detection using FBD.


Back to contents