Rapid detection and identification of plastic waste based on multi-wavelength laser Raman spectroscopy combining machine learning methods

Plastic waste has become a significant environmental concern, necessitating advancements in recycling efficiency. Enhancing the purity of recycled plastics facilitates the selection of suitable processing methods for different materials, thereby optimizing the recycling process. This study proposed a multi-wavelength laser Raman detection method and system to enable rapid and accurate identification of plastic waste. By analyzing the Raman spectra of various plastics under different laser wavelengths and introducing a fluorescence coefficient to quantify wavelength impact, the attribution of Raman characteristic peaks for distinct plastics has been elucidated, and the integrated area of Raman spectra across seven bands was identified as the key parameters for identifying plastics. By comparing neural networks, random forests, and k-nearest neighbor algorithms, it was determined that the k-nearest neighbor algorithm achieved the highest accuracy of 97.4 % and fastest identification speed of 1.2 ms/item when using integrated area of 7 characteristic bands as input. A plastic identification model incorporating data augmentation and k-nearest neighbors was finally developed and validated. A 100 % identification rate for actual waste plastic can be achieved by utilising a multi-wavelength laser Raman spectroscopy database. The results demonstrated that the multi-wavelength Raman system was highly effective for online or rapid recycling applications, enabling precise sorting of mixed plastic waste. This system significantly enhances the quality of recycled feedstock, contributing to the sustainability of plastic waste management.