Study On Detection Method And Recognition Algorithm Of Plastics Based On Raman Spectroscopy

In this paper,12 different kinds of plastic that included ABS,EVA,HDPE,PBT,PC,PET,PMMA,POM,PP,PPO,PP and PVC were taken as the object of the study.Raman spectroscopy was applied to the study on detection and recognition of plastics.The main research contents are as follows:Firstly,the collection method for plastic samples was established through the application of Raman spectroscopy.The four influence factors,including laser wavelength,accumulation time,confocal aperture and scanning range were optimized as following:the laser wavelength of 785 nm,accumulation time of 25s,confocal aperture 500μm and scanning range200-1800cm^-1.For black plastic samples,cryogenic sampling attachment,the liquid nitrogen refrigeration platform was used to help relieve the strong thermal effect which can destroy the sample.Through experimental analysis,when the laser wavelength of 785nm,laser power 10mw and accumulation time of 30s was the optimum acquisition condition.Then,the data of the collected plastic Raman spectra were processed,including spectral pretreatment and spectral feature extraction.For the spectral pretreatment,the polynomial least squares fitting method was used to smooth the spectrum and improve the signal-to-noise ratio.The number of polynomial was 1 and the window size was 3.The baseline correction method with line segment type baseline was carried out to eliminate the influence of fluorescence background and noise background.The maximum–minimum normalization method was applied to normalize the spectral data and eliminate the influence of the data size difference for the subsequent data analysis.Principal component analysis（PCA）was applied to extract spectral characteristics,the first 11 principal components which the cumulative contribution rate was 90.45%were selected as the new feature variab les for subsequent modeling analysis,which reduced the amount of data processing,simplified the processing difficulty of the problem.Finally,BP neural network model and support vector machine model were established with the processed spectral data.345 samples forming the training sample set was used to train the models,the optimal parameters of the model were determined by comparative analysis.For BP neural network model,the optimal network parameters were determined as follows:number of input layer node 11,number of hidden layer node 8,output layer node 4,learning rate 0.1,error 0.05 and training 1000 times.The support vector machine model had the best training effect when the data was processed with[-1,1]normalization method and took polynomial core function to training model.111 samples forming prediction sample set was used to test the reliability of the two models.The results showed that,for BP neural network model,the recognition rate of training sample set was 97.97%and the recognition rate of the prediction set sample was 94.60%,which basically met the requirement of plastic recognition rate.For support vector machine model,the recognition rate of the prediction sample set was 99.10%,which had a good recognition effect on plastic.The innovation points of this paper are as follows:1)Raman spectroscopy was applied to the study on recognition of plastics and a large number of sample were into the study and recognition models were established which provided new ideas for plastic recognition;2)With the combination of cryogenic sampling attachment,Raman spectroscopy was used to detect and identify black plastics.