The Ramachandran plot is among the most central concept for validating the conformation of protein structures, which plays an important role in structural biology. However, the favored regions defined by the traditional Ramachandran plot are too wide and contain inaccurate structures. For this lack, a method based on Support Vector Machine and Bayesian Optimization, SVM-Rama, is proposed to optimize and subdivide the definition of favored regions for the Ramachandran plot. The present study aims to improve the accuracy of the favored regions to specific secondary structure species of proteins and then to validate and annotate protein secondary structures simply and accurately. The results show that it has a high accuracy close to the best performance of traditional methods in secondary structure annotation but at lower training and computational costs than traditional methods do.