Image Forest Fire Segmentation Using Dirichlet Process Mixture Model

Abstract

Image analysis in the detection of forest fires aims to minimize damage by timely intervention to limit the damage. In this work, we will focus on image segmentation nonparametric Bayesian models used in forest fire detection. This model defines a probability distribution over functional spaces (of infinite dimension). We have therefore chosen to study the processes of Chinese restaurants in what is called mixture models. We used Gibbs sampling methods to generate the weights used in the representation of the Dirichlet process. Simulations were performed on signals, synthetic and real images, the results are promising.