Concepts et techniques d'integration du contexte spatial dans les modeles de ponderation des donnees multisources

The present research is concerned with the development of spatial analysis tools for multisource data modelisation applied to prospecting and planning problems under constraints. The main pursued objective was the explicit integration of the spatial dimension into a weighting process from multiple constraints associated with spatial entities. Several weight spatialization approaches are elaborated and the corresponding developed algorithms are experimented on the special case of finding a least-cost shortest path. First, a conceptual frame is used to define explicit rules of the spatial dimension integration. Next, two spatialization. approaches are defined. The first one is a global method where the weighting functions used take into account the frequency distribution of scores associated with spatial entities. The second approach consist in the calculation of a contextual weight for each spatial entity according to its neighbourhood. Two contextual methods have been suggested: an a posteriori approach applied on weights and an a priori approach applied on scores; the a priori approach applied to several neighbourhood levels constitute an adaptive weighting method. The a posteriori approach is studied by means of two segmentation algorithms whereas the a priori one is investigated by an integrated tool of segmentation and neighbourhood based on a Voronoi diagram structure. The adapted and developed algorithms were applied on artificial data cases and a real data case relative to the optimization of power lines under spatial constraints. The global weights spatialization, used on specific frequency distribution of scores, allows a significant reduction of a shortest path total cost compared to a conventional weighting process. The effect of the contextual a posteriori weighting method on a shortest path total cost is found to be dependent on the segmentation configuration. When we use this approach, the lowest total cost of a power line transmission line---resulting from a reduction of both the length and the scores costs of the path---is observed when the spatialization of weights is made within Voronoi polygons. The neighbourhood integration to the weighting process produce weights which vary on a wide range of values; this is basically due to the fact that regions having similar scores but dissimilar homogeneity are weighted differently. When the considered neighbourhood is sufficiently extended, the total cost of shortest path is reduced by comparison to the cost resulting from a non-contextual weighting. The spatialization of weights associated to planning constraints is found to be a tool for simulating the relationships and the particularities of the spatial entities.