Improvement of dry paper waste sorting through data fusion of visual and NIR data

Philipp Klippel, Matthias Zisler, Frank Schröder, Sebastian Schleich, Alla Serebryanyk, Claudius Schnörr

Abstract:

Near Infrared (NIR) spectroscopy is a well-known sensor technology which is used in many applications to gather information about chemical composition of materials. For paper waste sorting this information can be used to classify different paper classes which enables better
sorting and higher recycling quality. With a small number of NIR scores and assuming more or less unimodal clustered data, a pixel classifier can still be crafted by hand, using knowledge about chemical properties and a reasonable amount of intuition. Additional information can be gained by visual data. However it is not obvious how this information can be well captured by describing features, and what features are finally important for successfully separating the paper classes in feature space. Due to the huge variety of possible visual features, e.g. based on color, saturation, textured areas with different structure size, etc., a rigorous feature analysis becomes inevitable. We therefore have chosen a pattern recognition approach to deal with the curse of dimensionality. By exploiting a classification tree and a variety of additional visual features, followed by a forceful feature selection, we achieve a recognition rate of 78% for 11 classes, compared to 63% only using NIR features. The feature reduction shrinks the otherwise high computational burden to compute all features and furthermore even increases recognition rate slightly. While some visual features like color saturation and hue showed to be important, some NIR scores could even be
dropped.