In order to solve the problems of traditional characteristic spectral peak recognition and positioning methods, such as low recognition rate, large positioning error, and being unable to obtain spectral line-shape function, we propose a characteristic spectral peak recognition and positioning method based on an improved sine cosine algorithm. A sine cosine algorithm is improved by dynamic conversion probability, and then combines with the fitting methods of various spectral line-type functions (Gaussian, Lorentzian, and Voigt). The corresponding positions of characteristic spectral peaks can be obtained by iterative optimization. The improved method can not only locate characteristic peaks precisely, but can also obtain the line-type function of the spectrum. The experiments show that the proposed method significantly improves the recognition rate, positioning accuracy, peak fitting effect, and noise suppression ability for strong, weak, and overlapping peaks.