In view of the fact that the current flame image edge extraction algorithm was susceptible to the influence of approximate brightness, make the edges not clear enough, integrity was not strong, etc, an image edge extraction scheme based on local binary mode coupled with double threshold LM optimization was defined. Firstly, the color image was transformed into gray image, and the gray level of the image was adjusted according to the statistical distribution of the image. Secondly, the image was smoothed by using the Gaussian filter to eliminate the influence of noise. Then, local binary patterns (LBP) were used to process images, and global threshold technique was used to calculate the local features of edges. Then, the non-maximum suppression operator was used to get more accurate edges, and two thresholds are selected to create two different edge images in non-maximum suppression. In order to speed up the determination of non-maximum suppression thresholds, Levenberg-Marquardt (LM) optimization algorithm was used to optimize the cost function based on mean square error, eliminating false edges while retaining small edges. In addition, the area, average value, standard deviation and variance of the flame image were used to analyze the flame image, thus the flame temperature can be accurately obtained, and the stability of flame combustion can be predicted. Experiments show that compared with the current flame image edge detection technology, this proposed algorithm has higher edge detection quality and better flame edge integrity, which can remove noise and irrelevant artifacts.