Segregating wood wastes by repetitive principal component analysis of near infrared spectra

Abstract

To improve the recycling rate of wooden materials, it is necessary to classify wood waste by disposal method and usage. In the industrial manufacture of these materials, rapid and accurate determination of their chemical and physical properties is critical for a stable supply of wood products with reliable quality. In this study, we investigated a discriminant analysis process for waste wood products using hyperspectral imaging with a newly developed repetitive principal component analysis. Hyperspectral images of four types of wood waste (plywood coated with resin, preservative-treated wood, hardwood and softwood) were acquired. The mean spectrum of each sample was extracted from a hypercube in order to build a classification model. A novel classification method based on principal component analysis, named repetitive principal component analysis, was developed. A total of three repetitions of principal component analysis were performed to classify the four types of wood waste. Cross-validated results of repetitive principal component analysis resulted in classifications greater than 85% for any of the four wood waste types. The discriminant model was then applied to single-pixel spectra of the hypercube to form a prediction map. Hyperspectral imaging, with the aid of the new repetitive principal component analysis discriminant analysis, is a powerful tool in wood recycling processes.