Study On Identification Of Mixed Plastic Wastes Based On Near Infrared Spectrum

“Urban mineral” is known as “the21st century sunrise industry” under thebackground of shortage of resources and energy. However, it is common in thedevelopment process of urban minerals that there are a series of problems such as therecycling utilization rate of mixed plastic waste resources being low, the recycledproducts being low additional value, and so on. The reason is that a variety of mixedplastic wastes recycled as a hybrid state make it difficult to be accurately divided intoa single type of plastic. Therefore, it has become urgent to study an accurateidentification and separation technology industrially to recycle mixed plastic wastes.This paper mainly analyzed the sources and types of the mixed plastic wastes,the current status on recycling utilization of obsolete plastics in the domestic andoverseas, conventional identification separation and recycling methods. And chemicalmetrology methods used in near infrared spectrum analysis technology were alsoreviewed. On this basis, this paper proposed a separation process of mixed plasticwastes and analyzed the difficulties and the main technical problems needed to besolved in that process.Secondly, with respect to studies on pretreatment methods of near infraredspectra of plastic wastes including ABS, PET, PVC, PP, PS and PE, this paper adoptsthe Savitzky-Golay least-square fitting derivation combining the wavelet analysismethod to eliminate baseline drift and filter out the spectral noise. Comparing the firstderivative spectra of6kinds of plastics with different window sizes and the evaluationresults of root mean square error and arithmetic average, it turned out that the optimalparameters are3polynomial and13window sizes. And the effects of wavelet smoothfiltering were evaluated by SNR. When the first derivative spectra is decomposedunder the scale of2, and the sym17wavelet generating function,“rigrsure” thresholddetermination method, threshold reset method of “sln” and soft threshold methodwere used, the de-nosing effect was better than that of other methods.Thirdly, the modeling method based on near infrared spectra of mixed plasticwastes was studied. It was principal component analysis for calibrating samples in thefirst place, then selecting characteristic wavelength for Fisher discriminant analysisand seting up canonical discriminant functions. Testing the stability of the model by means of calibration sample set itself validation and cross validation, the rates ofreturn to discriminant model are100%and84.9%, respectively. While thediscriminant accuracy is100%when unknown plastics were used in the externalvalidation of fisher discriminant model, which makes the identification of mixedplastic wastes come true.Finally, through comparing the differences between the relative reflectance ofdifferent characteristic wavelengths, another algorithm model to identify six kinds ofplastic wastes was established. The discriminant accuracy of the calibrating samplesreached97.3%, and the prediction accuracy of the unknown samples was93.5%,which proves that this method can also accurately identify the six kinds of plasticwastes.