de Carvalho, Luis
5 EXPERIMENTAL RESULTS
By multiplying the wavelet images, all directions of changes
get positive values. Visualisation can be readily done with a
simple colour composite. To visualise changes from dark to
light (e.g. deforestation in TM band 5), one must use the
oldest image to make the composite (Figure 5). Changes from
light to dark (e.g. reforestation in TM band 5) are better
visualised when using the most recent image. This is because
the background at sites where changes are to be visualised
must be dark so that the changes of interest could be
emphasised. Ground-truth for changed areas was acquired
during field trips in 1999 and from interpretation of aerial
photographs from 1979 and 1984. All changes detected by
this visualisation procedure did occur although their
quantification was not possible. Note that in Figure 5 the
whole triangular forest fragment at the bottom right
disappeared between 1985 and 1998 although only its centre
is being enhanced.
Figure 5. Landsat TM from 1985. Changes of
interest coloured with red, band 4 with green and
band 3 with blue. Arrows indicate areas where (a)
deforestation and (b) mining increased.
In the 3D graphs of Figure 6 note that lots of unimportant information that could be considered noise does not appear in
the change image built with the combination of wavelet images (Figure 6b). This ‘cleaning’ effect makes the analysis
and understanding of remotely sensed images or outputs from image processing much easier.
Figure 6. (a) Difference image between band 3 of Lansat TM from 1998 and band 2 of Landsat MSS from 1981 with
respective magnitude graph. (b) Product of differences at second and third scales and respective magnitude graph.
Deforested areas (red arrows), reforested areas (blue arrows) as well as new mining sites (yellow arrows) were
successfully pinpointed in Figure 6b without previous radiometric rectification or threshold definition while differences
not related to land cover changes were bypassed. Verification was done by visual evaluation of ancillary data.
344 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.