Automatic multistage classification system for plastic bottles recycling

In this work, an artificial intelligent plastic bottles classification system is proposed, developed and tested. Classifying bottles based on their chemical composition and color is attempted. Near infrared (NIR) reflectance measurements are used to identify bottle composition class. Charged coupled device (CCD) camera with the fusion of quadratic discriminant analysis (QDA) and tree classifiers are used to detect the bottle color.Results have shown that the dip wavelength and average values of the reflective NIR spectrum could be used as features to distinguish between chemical compositions. This resulted in 94.14% classification accuracy. In addition to various preprocessing techniques, the use of principal component analysis algorithm for bottle orientation facilitates the detection of the bottle color avoiding mixing it with the bottle's label or cap. Ninety-two percent color classification accuracy is achieved for clear bottles while 96% is achieved for opaque one, with proposed method. The aggregate classification accuracy of the combined system (i.e. accurate classification of color as well as chemical composition) is 83.48%.