Material flow composition dataset: RGB-camera-based single-particle images and particle mass data of shredded WEEE

Malte Vogelgesang; Alice do Carmo Precci Lopes; Emanuel Ionescu; Liselotte Schebek

doi:10.1016/j.dib.2025.112413

Material flow composition dataset: RGB-camera-based single-particle images and particle mass data of shredded WEEE

Data Brief. 2025 Dec 22:64:112413. doi: 10.1016/j.dib.2025.112413. eCollection 2026 Feb.

Authors

Malte Vogelgesang¹, Alice do Carmo Precci Lopes^{1

2}, Emanuel Ionescu¹, Liselotte Schebek²

Affiliations

¹ Fraunhofer Research Institution for Materials Recycling and Resource Strategies IWKS, Brentanostr. 2a, Alzenau D-63755, Germany.
² Institute IWAR, Technical University Darmstadt, Franziska-Braun-Str. 7, Darmstadt D-64287, Germany.

Abstract

Mass-based material flow compositions (MFCOs) are among the most relevant indicators for the quality of sorting processes in waste management. They provide the basis for the assessment and improvement of the purity and yield of material streams. While MFCOs are currently primarily determined through labor-intensive manual sorting analyses, recent research has explored sensor-based material flow characterization (SBMC). Near-infrared spectroscopy (NIR) is effective for plastic waste but cannot reliably distinguish many material types in shredded waste electrical and electronic equipment (WEEE). Therefore, this article aims to provide data for research on SBMC with more versatile and often already implemented RGB cameras in sorting plants. The generated data contain shredded WEEE, which was recorded on a sensor-based sorting machine on an industrial scale. The samples were taken from an industrial WEEE recycling facility in Germany and manually presorted into four main material types. From these sorted materials, different subsamples were created according to each dataset's objective. The three datasets consist of cropped RGB line-scan camera images and mass measurements. Dataset 1 contains labeled single-particle images for the training of image classification, object detection, and segmentation networks and is organized by material type (ferrous, non-ferrous, printed circuit board, and plastic) and by particle size (12.5 mm - 25 mm and 25 mm - 50 mm). Dataset 2 follows the same structure and provides single-particle images together with per-object masses for the training of regression models for particle mass prediction. Dataset 3 provides single-particle images from three mixed samples with ground-truth compositions to validate models trained on Datasets 1 and 2. To serve as an additional validation dataset, Dataset 1 also includes the mass of each material fraction. These datasets enable the development and evaluation of models for the identification of material types in shredded WEEE, for particle mass prediction, and sensor-based material flow characterization, supporting improved process monitoring, quality control, and optimization of sorting processes.

Keywords: Computer vision; E-waste; Machine learning; Material flow characterization; Particle mass prediction; Recycling; Sensor-based sorting; Waste electrical and electronic equipment.