Recognition Technology Of Raman Spectrum Based On Deep Learning

As an accurate and efficient material identification technique,Raman spectroscopy has the advantages of simple operation,high sensitivity,no damage to samples and little interference from fluorescence signals,it is widely used in industry,agriculture,biology,medicine and so on.In the classification and recognition of Tunku Abdul Rahman spectrum,the machine learning algorithm is usually used,but the analysis process of traditional machine learning algorithm is relatively complicated,so Tunku Abdul Rahman spectrum needs to be extracted artificially,the method of artificial feature extraction relies on a wide range of professional knowledge and prior knowledge,and the artificial feature extraction will cause the loss of spectral information,and then affect the accuracy of the analysis results.In recent years,deep learning has become a hot research topic,and convolutional neural network,as a classical algorithm in deep learning,has achieved good results in image classification and recognition,this provides an idea for studying a new classification and recognition model of Raman Spectrum.According to the experience of convolutional neural network in image recognition and the characteristics of Raman spectrum signal,a one-dimensional convolutional neural network suitable for Raman spectrum classification and recognition is designed in this paper:First of all,the research status of Raman spectroscopy and deep learning is introduced,and the emphasis is put on the research status of convolutional neural network in Raman spectroscopy identification,then it explains the traditional machine learning classification algorithms and the convolutional neural network.Then,the Raman Spectra data of three kinds of edible flavors(vanillin,methyl vanillin and Ethyl Vanillin)were collected,and the collected Raman Spectra were analyzed by characteristic peak analysis and spectral pre-processing,according to the training requirements of the convolutional neural network,three methods of Raman Spectral data enhancement are designed to obtain enough Raman spectral data and ensure the diversity of the spectral data,in order to complete the comparison between the convolutional neural network classification algorithm and the traditional machine learning classification algorithm,and verify the dependence of the two classification algorithms on the spectral preprocessing,three different data sets of Raman Spectra were made.Finally,according to the characteristics of Raman spectrum signal,three kinds of one-dimensional convolutional neural network models for Raman spectrum classification are designed and compared.In order to test the advantage of the convolutional neural network in dealing with Raman spectral recognition problem,the modeling analysis process,the analysis accuracy,and the robustness of the model were taken into account,convolutional neural network compares ml with traditional ml algorithms.