Plastic identification using NIR spectral analysis and KNN model with correlation feature selection

The increasing demand for high-purity postconsumer plastics for recycling highlights the importance of efficient material identification. Near-Infrared (NIR) spectroscopy is widely used to capture unique spectral fingerprints of materials for identification purposes. However, developing efficient and lightweight machine learning models with NIR spectral data for accurate and real-time classification of plastic materials is still in high demand. This paper presents a new hybrid approach combining the k-nearest neighbours (KNN) model with the filter-based correlation feature selection (CFS) technique for plastic identification. Experiments were conducted to test six key recyclable plastic polymers consisting of Polyethylene Terephthalate (PET), High-Density Polyethylene (HDPE), Low-Density Polyethylene (LDPE), Polyvinyl Chloride (PVC), Polypropylene (PP), and Polystyrene (PS). Feature selection techniques - CFS and principal component analysis (PCA) are applied to extract the essential discriminative features of each type of plastic. With the extracted features, KNN and support vector machine (SVM) models are developed respectively for plastic identification. Experimental results demonstrate that the CFS-KNN model achieves a success rate of 100% with a computational time of 40ms under laboratory conditions, outperforming the PCA-SVM, PCA-KNN and CFS-SVM models in terms of accuracy and computational efficiency.