Waste Material Classification: A Short-Wave Infrared Discrete-Light-Source Approach Based on Light-Emitting Diodes.

Waste material classification is a challenging yet important task in waste management. The realization of low-cost waste classification systems and methods is critical to meet the ever-increasing demand for efficient waste management and recycling. In this paper, we demonstrate a simple, compact and low-cost classification system based on optical reflectance measurements in the short-wave infrared for the segregation of waste materials such as plastics, paper, glass, and aluminium. The system comprises a small set of LEDs and one single broadband photodetector. All devices are controlled through low-cost and low-power electronics, and data are gathered and managed via a computer interface. The proposed system reaches accuracy levels as high as 94.3% when considering seven distinct materials and 97.0% when excluding the most difficult to classify, thus representing a valuable proof-of-concept for future system developments.

Material classification based on a SWIR discrete spectroscopy approach.

A crucial yet difficult task for waste management is the identification of raw materials like plastic, glass, aluminum, and paper. Most previous studies use the diffused reflection spectroscopy for classification purposes. Despite the benefits in terms of speed and simplicity offered by modern compact spectrometers, their cost and the need for an external, wide-spectrum source of illumination create complications. To address this issue, the present paper proposes a discrete spectroscopy method that utilizes short-wave infrared (SWIR) reflectance to identify waste materials, exploiting a small set of selected wavelengths. This approach reduces the complexity of the classification data analysis and offers a more practical alternative to the conventional method. The proposed system comprises a single germanium photodetector and 10 different light emitting diodes (LEDs). The LED wavelengths are selected to maximize the system sensitivity towards a set of seven different waste materials. Using a classification strategy relying on support vector machines, the proposed methodology reaches a classification accuracy up to 98%.