Abstract: The correct recognition of flame combustion state is a prerequisite for maintaining stable combustion.However,the existing methods have a low intellectual level,strong subjectivity and difficulty in the recognition of the combustion state.Therefore,in order to improve the accuracy of recognition,this paper proposes a flame image recognition method based on tensor decomposition.According to the distribution characteristics of the flame pixels in the RGB color space model,the ascending dimension of the image sequence is constructed into a three-dimensional tensor,and,on this basis,the parallel factor analysis method is used to identify the flame images.The analysis results show that,compared with the traditional regional recognition and feature fusion method of single flame images,the flame image tensor is decomposed directly,which not only finds the relationship between the flame image sequences,but also retains more information to a greater extent to improve the accuracy of image recognition.Compared with the theoretical value,the average error of the new method is within 0.5%,and the accuracy is more than 99.5%.It provides reference for continuous flame image recognition.