The TIR (Thermical InfraRed) channel owns wave lengths sensitive to the emission of heat. So the higher the heat of an area, the higher the intensity of the corresponding pixel of the image. Then the forest fire can be caracterize by pic intensity on that kind of images. We present a fully automatic method of forest fire detection in satellite images based on the random field theory. First we model the image by a realization of a Gaussian field. The fire areas which have high intensity and are supposed to be a minority are considered as foreign elements of that field : they are rare events. Then we determine by a statistical analysis a set of probabilities which caracterizes a degree of belonging to the Gaussian field of a small area of the image. By complementarity, we estimate the probability that this area is a potential fire.
