Abstract: Computer vision is one of the popular topics in computer technology research field,and the selection and extraction of target object are the core issues of computer vision．The article put forward the method to sieve the assemblage features of flame image from three aspects—textures,dynamic and geometric,based on analyzing and researching the method to predict the flame,and specifically introduced the extraction method of dynamic features．The article made a more in-depth study on the feature selection of flame recognition,made comprehensive utilization of the content features of the flame,and proposed the new fitness function based on the genetic algorithm,and made the feature selection more scientific,while SVM can make full use of its advantages in identification process based on the theory of structural risk minimization,so it achieves the desired effect in the experiment.

Key words: Feature selection,Genetic algorithm,Support vector machine,Classification,Forest fire