We used the additional source ``card of the favourable areas'', highly reliable, to set up group K₁. The other images (spectral bands) are sources less reliable than the preceding one. This is why we choose to gather them in group K₂.

1.

Before performing the fusion of information of the groups K₁ and K₂, it is essential to perform a fusion of the sources composing each group, in order to obtain a single possibility distribution for each group. As group K₁ is only composed of one source (the card of the favorable zones), no fusion is necessary for this group. On the other hand the group K₂ gathers several sources (four spectral bands). The fusion of the information provided by these sources must be performed to obtain a single possibility distribution for the group K₂. Quantified adaptive fusion has been chosen as it produces the most interesting results.

2.

We consider now the phase of fusion of the possibility distributions of the groups K₁ and K₂. At this level, it is necessary for us to solve the problem of the calculation of the height of the intersection between the distributions and . As in the preceding cases, and for the same reasons, we propose to calculate this degree locally to each pixel, but globally for each class. So:

(76)

   

  

Table 38: Fusion by priority of groups K₁ (``card of the areas favourables to each class'') and K₂ (the 4 spectral bands MSS4 to MSS7).

Rates of classification obtained
Class	Number of pixels   correctly classified (A)	Number of pixels in samples      (B)	Rate of pixels     correctly classified (A/B)
1	430	459	93.68%
2	406	459	88.45%
3	295	306	96.41%
4	368	391	94.12%
5	378	459	82.35%
6	430	459	93.68%
7	440	459	95.86%
8	448	459	97.60%
9	443	459	96.51%
TOTAL	3638	3910	93.04%

Fusion by priority of the groups K₁ and K₂ (definite above) leads to a good result with an average of 93.04% of the pixels of the samples correctly classified. Additional information ``card of the favourable zones'', highly reliable, was correctly put forward, compared to the other sources. But the spectral information contained in the spectral bands played a significant role of refinement of information.

The image obtained (table 38) presents an aspect more disturbed than the image of the theoretical distribution ``ideal'' of each class (table 37). This disturbed image is much closer to the physical reality presented by the information of the image LANDSAT than the ``ideal'' card calculated.

Classes 1 (low forest sempervirente) and 4 (semi-deciduous forest or dry deciduous), over-represented on the ``idea'' card, yielded their place mainly to class 6 (mosaic of shrubby savanna and reforestations).