Study On Pattern Recognition Of Classification Of Near Infrared Spectroscopy In Plastic Sorting

With the continuous progress of industrialization,the production scale of the global plastics industry has expanded rapidly.Plastic products are widely used in various industries such as construction,transportation,home appliances,and agriculture,which greatly facilitates people's production and life.However,the life cycle of plastic products is relatively short,and a large amount of waste plastics have been accumulated around the world.In order to improve the environment and save resources,research on the recycling of waste plastics is an urgent need in today's society.The identification and classification of waste plastics is the first step in the plastic recycling process and the key to ensuring the quality of plastic recycled products.The identification and classification of waste plastics is the first step in the plastic recycling process and the key to ensuring the quality of plastic recycled products.Therefore,this paper takes waste plastic as the research object,combining near infrared spectroscopy technology with pattern recognition and conducts research on the recognition and classification technology in the plastic recycling process.The main research contents are as follows:(1)Near infrared spectroscopy pretreatment and extraction of characteristic wavelength of plastic samples.The Monte Carlo cross-validation method was used to remove abnormal samples.The Monte Carlo cross-validation method is used to remove abnormal samples.The moving average smoothing(MAS)and Savitzky-Golay(SG)convolution smoothing were used to smooth the spectrum,and leave 1 the cross-validation accuracy rate was the basis for selecting the smooth window size.In order to remove the redundant information of the plastic spectrum data and improve the modeling speed,Principal component analysis(PCA),successive projections algorithm(SFA),competitive adaptive reweighted sampling(CARS)anduninformative variable elimination(UVE)were used to extract characteristic wavelength for the original spectrum and pre-processed spectrum.(2)The establishment of pattern recognition classification model.BP neural network,support vector machine(SVM)and partial least squares discriminant analysis(PLS-DA)were used to establish the pattern recognition classification model,and genetic algorithm(GA)was used to optimize the weight and threshold of the BP neural network.The grid search method(GSM)and particle swarm optimization(PSO)were uesd to optimize the penalty factor and kernel function parameters of the SVM algorithm.Combining the principles of modeling methods,the classification accuracy was used as the evaluation index of the model.The best model based on BP neural network was S-G convolution smoothing + SNV + CARS+GA-BP.The accuracy of the training set was 99.212%,and the accuracy of the prediction set was 96.364%.The best model built based on SVM was S-G convolution smoothing + MSC + PCA + PSO + SVM.The accuracy rate of the training set was99.213%,and the accuracy rate of the prediction set was 95.152%.The best model based on the PLS-DA algorithm was S-G convolution smoothing + PCA +PLS-DA.The accuracy rate of the training set was 85.827%,and the accuracy rate of the prediction set was 82.424%.(3)The design of overall structure of the plastic sorter.Feeding device,conveying device,identification device and injection separation device in the plastic sorting machine were designed,using Solidworks software to carry out three-dimensional modeling of the above four devices.At the same time,the transmission device was improved,and the belt transmission was changed to ball transmission,which helped reduce the overlap of plastic samples and improve the sorting accuracy.