No abstract

International audienceCrisis management challenges decision support systems designers. One problem in the decision marking is developing systems able to help the coordination of the different involved teams. Another challenge is to make the system work with a degraded communication infrastructure. Each workstation or embedded application must be designed such as potential decisions made trought other workstations are treated as eventualities. We propose in this article a multi-agent model, based on an ant colony optimization algorithm, and designed to manage the inherent complexity in the deployment of resources used to solve a crisis. This model manages data uncertainty. Its global goal is to optimize in a stable way fitness functions, like saving lives. Moreover, thanks to a reflexive process, the model manages the effects of its decisions into the environment to take more appropriate decisions. Thanks to our transactional model, the system takes into account a large data amount and finds global optimums without exploring all potential solutions. In perspective, users will have to define rules database thanks to an adapted graphical interface. %Each rule associates, for each potential event, a goal with its fitness functions, and a list of possible tasks to do. Then, if the nature of the crisis is deeply unchanged, users should be able to change rules' databases

The segmentation of retinal vasculature by color fundus images analysis is crucial for several medical diagnostic systems, such as the diabetic retinopathy early diagnosis. Several interesting approaches have been done in this field but the obtained results need to be improved. We propose therefore a new approach based on an organization of agents. This multi-agent approach is preceded by a preprocessing phase in which the fundamental filter is an improved version of the Kirsch derivative. This first phase allows the construction of an environment where the agents are situated and interact. Then, edges detection emerged from agents' interaction. With this study, competitive results as compared with those present in the literature were achieved and it seems that a very efficient system for the diabetic retinopathy diagnosis could be built using MAS mechanisms.