Aiming at the problem of lack of mathematical quantitative evaluation index in band selection and identification index of existing high temperature target recognition method, a forest fire in Daxing'anling area of Inner Mongolia is taken as an example. Landsat8 OLI remote sensing image data are used to study the multispectral identification index of forest fire target. It is found that the "Separable Measure" index based on the variance analysis principle can be used as a mathematical quantitative evaluation index for the selection of the band in the high temperature target recognition method and the qualitative analysis method of the previous identification method. In the process of distinguishing between forest fire targets and other objects, the mathematical quantitative evaluation indexes all have the physical meaning of remote sensing target recognition and classification, and its classification suitability plays an important role in the identification and extraction of forest fire targets. After verification and analysis, the appropriate fire point identification index selected by the "Separable Measure" index is effective and accurate in the actual classification, which is consistent with the theoretical results.