Aiming at the weight difference between strong and weak peaks and the interference of noise peaks in the traditional reverse matching method (RMM), an improved reverse matching method (IRMM) is proposed in this paper. In this method, the weight attenuation function is introduced to optimize the weight proportion relationship between the strong peak and the weak so that the weight of each feature peak in the spectrum is distributed in a reasonable range, which avoids the situation that the weight of the strong peak masks the weak. Moreover, this method realizes the adaptive filtering of noise peaks by the method of dynamic noise filtering of the probability distribution function, which improves the recognition performance of the reverse matching method. In the experiment, many conventional Raman and surface-enhanced Raman spectra were used as verification samples, which were identified and verified based on a large database of conventional Ramanand surface-enhanced Raman. Experiments show that this method (IRMM) has a comprehensive accuracy rate of 91.52% under a large amount of data testing, which is greatly improved compared to the hit quality index method (HQI, 51.08%) and the traditional reverse matching method (RMM, 16.57%).