Research On Raman Spectroscopy Analysis Method For Mixture Component Detection

Mixture component recognition has been widely used in drug detection,material science and other fields.Raman spectroscopy has become an important tool for mixture component recognition because of its accuracy,rapidity and nondestructive advantages.However,the current Raman spectroscopy-based material component detection methods still have the following deficiencies in recognition accuracy and recognition speed:(1)The external environment,measuring instruments and the interaction between components make the Raman peaks of real components in the mixture shift to varying degrees,When the overlap of spectral peaks of the mixture is serious or the concentration of components is low,it becomes extremely difficult to extract the characteristics of spectral peaks,which has a great impact on the recognition accuracy of the mixture;(2)Under the environment of large-capacity database,the mixture recognition method based on spectrum database search strategy needs a long recognition time,which cannot meet the real-time requirements of the actual detection scene.To solve the above problems,this paper proposes some new methods for mixture component recognition.The specific research contents are as follows:1.Aiming at the problem that the peak shift of related components in the mixture spectrum brings great interference to the recognition accuracy,a mixture component recognition method based on incremental database search is proposed.This method uses historical detection data to construct a mixture database of known components for auxiliary search.By extracting the spectral peak characteristics of the related components of the known mixture,the similarity calculation error caused by Raman spectrum peak shift is compensated to improve the recognition accuracy.Using the proposed method to identify 81 mixtures,the recognition results show that the incremental database search method proposed in this paper can significantly improve the recognition effect of mixtures compared with the pure substance database search,and the recognition accuracy is increased from 76.34% to 92.83%,which verifies the effectiveness of the method.2.Aiming at the problems of low recognition efficiency and real-time inability to meet the detection requirements in large-capacity database scenes,a fast detection method of mixture components based on Raman peak feature clustering is proposed.Firstly,the spectral peak parameters of pure substances are transformed into clustering feature vectors.In order to eliminate the interference caused by the peak intensity to the clustering effect,the Raman shift and full width at half maximum of the spectral peaks are used as the main characteristics of the clustering.The Kmeans + + method is used to cluster pure substances,and the elbow rule is used to divide the pure substance database into several clusters.In the process of mixture recognition,the distance between the substance to be recognized and various clusters is calculated,and the clusters whose distance is less than the set threshold are selected for search and matching.The recognition results show that the recognition time of the proposed method is reduced by one time,and the recognition accuracy is not affected.3.Aiming at the problem that the mixture component identification method based on characteristic peak comparison cannot meet the detection requirements when the concentration of the substance to be identified is low or the interference of overlapping peaks is serious,a mixture component detection method based on depth learning is proposed.In this method,the original Raman spectrum of the mixture is transformed by continuous wavelet to highlight the characteristic changes of Raman peaks of the mixture at different scales Automatic extraction and recognition of mixture component features by deep convolution neural network;In addition,an Inception module with cascade convolution kernels of different sizes is introduced to enhance the information extraction ability of multi-scale characteristic peaks.The experimental results show that the introduction of Inception module and continuous wavelet transform can effectively improve the recognition accuracy of components,and the recognition accuracy of the six substances reaches 97.66%,95.9%,96.48%,98.82%,100%,and 96.52%,respectively.