Machine learning for waste segregation: Chinese researchers couple spectroscopy with sorting algorithms
28 Jul 2023 --- In a bid to accurately and reliably classify recyclable waste, researchers at Hefei University of Technology in China have developed a laser-induced breakdown spectroscopy (LIBS) technology. It can identify and classify recyclable waste into six categories: Paper, plastic, glass, metal, textile and wood.
Their study, published in AIP Advances, defines LIBS as an atomic emission spectroscopy technology that detects the sample’s elemental composition based on emission spectra. It is not affected by the ambient environment and light, samples’ shape and color.
“We have used LIBS technology for the first time to identify and classify recyclable waste,” says author Lei Yang. “This method has accurate, reliable, fast detection results and can achieve automatic detection.”
Identification and classification system for recyclable waste (Image credit: Lei Yang).Identifying spectra
The researchers employed machine learning models to advance the identification process. They collected the spectra of 80 recyclable waste samples and classified them through LIBS spectra.
The scientists also subclassified metal into iron, stainless steel, copper and aluminum and plastic into polyvinylchloride (PVC), polyoxymethylene (POM), acrylonitrilebutadiene-styrene (ABS), polyamide (PA), PE and polytetrafluoroethylene (PTFE) for recycling and reuse purposes.
Among the explored models, the combination of linear discriminant analysis (LDA) and random forest (RF) emerged as the most optimal for classifying recyclable waste. Additionally, a combination of principal component analysis and RF was deemed the most effective for subclassifying metals and plastics.
“What surprised us the most was that by using LIBS technology for classification and recognition without any preprocessing of the waste object, the results are satisfactory,” Yang remarks.
Researchers found the model of LDA with RF in classifying recyclable waste achieved an accuracy of 100%. For subclassifying metals and plastics, the PCA(9D) + RF model scored an accuracy of 99.53%.
Waste management
According to the authors, their findings indicate the potential of LIBS technology in improving recycling efficiency and waste management practices.
“The LIBS technology combined with drop-dimension algorithms and machine learning algorithms can realize high-precision identification, classification and subclassification for recyclable waste, which provides a new automatic real-time detection method and technology in the environmental protection field,” state the researchers.
Furthermore, they highlight that waste management is a “severe social issue today and has been on a steady rise.”
The researchers plan to build on the results of their study by increasing the number of waste samples and incorporating other forms of waste, such as kitchen waste. They hope to deepen the understanding of transparent glass detection with LIBS, opening new avenues for recycling and waste management.
“Most identification and classification of waste methods are based on the differences in their physical properties, such as gravity separation, magnetic separation, electrostatic separation and eddy current separation. These methods are simple and can only detect a few categories of waste, and the classification results are inaccurate and cannot be used for waste reuse.”
“With the development of image processing and spectroscopic techniques, more categories of waste can be identified and classified by obtaining their physical and chemical information,” they conclude.
Edited by Radhika Sikaria
