Target Detection and Identification using neural networks and multi-agents systems Lieutenant R. Cozien, C. Rosenberger, P. Eyherabide, J. Rossettini, A. Ceyrolle French Department of Defence Ecoles militaires de Saint Cyr Coëtquidan Centre de Recherche – équipe Informatique & Simulation 56381 Guer Cedex phone : +33 2 97 73 50 30 fax : +33 2 97 73 50 83 Email : jarod@mailhost.esm-stcyr.terre.defense.gouv.fr Abstract - Our purpose is, in medium term, to detect in air images, characteristic shapes and objects such as airports, industrial plants, planes, tanks, trucks, … with great accuracy and low rate of mistakes. However, we also want to value whether the link between neural networks and multi-agents systems is relevant and effective. If it appears to be really effective, we hope to use this kind of technology in other fields. That would be an easy and convenient way to depict and to use the agents' knowledge which is distributed and fragmented. After a first phase of preliminary tests to know if agents are able to give relevant information to a neural network, we verify that only a few agents running on an image are enough to inform the network and let it generalize the agents' distributed and fragmented knowledge. In a second phase, we developed a distributed architecture allowing several mutli-agents systems running at the same time on different computers with different images. All those agents send information to a "multi neural networks system" whose job is to identify the shapes detected by the agents. The name we gave to our project is Jarod. Keywords: multi-agents, neural network , TCP/IP communications, image processing. 1 Introduction Image processing is certainly a major application of neural networks - NNs - but, above this use, our aim was to know how agents can become the eyes, the hands, … the sensors, if not, the senses of NNs, which are very often some kind of brains in a box : blind, one-armed and legless. Many neural architectures and algorithms have been designed, but very often, a specific architecture suits a specific need and rarely many differents needs. Military needs are also quite specific, indeed, we know what we are looking for, but we rarely know when we shall find it, neither the kind of picture holding the information, and whether the quality level of the picture is going to be satisfactory. As we can't code a new architecture each time the military situation is evolving, we tried to find a technology allowing in the same frame many different ways to extract the edges and to identify them. This approach may also find some roots in Minsky theory [4] where "agents" act together in order to fulfill a particular task which is out of the range of any single agent. For us "agents" are viewed as a triple ability : perception, decision making, action. (The action ability includes the possibility to communicate with other agents). If we reduce the middle phase to a minimum, like choosing a direction or objects to move, … agents can be described as reactive entities perceiving then modifying their environment. We talk about multi-agents systems - MAS - because these systems are made of numerous identical agents or groups of identical agents. Those systems shall be used in many cases, but mainly with a view to simulate real phenomenons. We can surely use more complex agents in terms of decision making, but "multi- agents programming" means for us numerous identical reactive agents, otherwise, we talk about "agents programming". In some cases we can observe particular phenomenons gathered under the label of "emergence". This means that some MAS develop the cooperation between agents with no central coordination on purpose to accomplish a task that no single agent would reach on its own. But nothing really formal and convincing as been written on it yet. Thus, there is no "multi agents means" to know and to use the knowlegde of the whole system. This knowledge, if it exists, is diluted in the agents, and worse, the