This study focuses on identifying and detecting spectral molecular bands caused by changes in internal energy levels of molecules, which contributes to the research of stellar spectral types and parameter estimation. First, considering the curve trend of molecular bands, pseudo molecular bands that have a W shape but an obvious downward trend should be eliminated by using curve analysis to identify molecular bands. Bringing the identification idea of multi-type and multi-classification criteriainto the model, the four parameters of the detected molecular band peak depth, W-shaped width, curve trend, and rebound trend are adoptedas training features, which consider comprehensivelythe change rate of starting point, change trend of curve, extreme point distribution and the factors of curve shape. Secondly, the LightGBM (Light Gradient Boosting Machine) model is used to identify the spectral and molecular band characteristic parameters of F-type stars with an accuracy of 97.62% and 99.16%, respectively, to verify the feasibility and reliability of this method. This work can not only excavate the late stars and improve the accuracy of data labels but also automatically identify the late stars by using the LightGBM machine learning model to detect the unknown spectrum based on accurate recognition, which improves recognition efficiency and reduces memory occupation.