TwinSpectra Dataset

TwinSpectra Dataset is a novel collection of two sets of spectroscopic measurements of refuse-derived fuel particles sampled during the cement manufacturing process. The measurements were taken during two separate sessions using two distinct devices: The first session is delivered with the measurements_2151.csv file. It contains 718 spectroscopic measurements with particle group labels and subset folds. It provides 2151 reflectance columns (data features) spanning 350 to 2,500 nm. It was collected using the ASD FieldSpec 4 Hi-Res. The second set we put into measurements_125.csv file. It holds 720 spectroscopic measurements with 125 data columns, associated with folds and labels, similar to the first file. The data was acquired using Viavi microNIR, which captures the measurements in the range of 908.1 to 1,676.2 nm. The TwinSpectra Dataset aims to enable the development of remote sensing detection algorithms. The particle group information (group column) can be used to train machine learning classifiers for six different classes: Foam, Foil, Rubber, Textile, PC or 3D (for paper or cardboard and three-dimensional samples, for the last two classes). Additional data on subfolds (fold column) could support a cross-validation procedure for evaluating the algorithms. This package includes examples of Python Jupyter notebook files. We provide two files: one for basic measurement analysis and preview, and one for preparing the initial baseline machine learning model. More details can be found at: https://doi.org/10.1016/j.dib.2026.112920 (Data in Brief paper) Keywords spectroscopy, RDF classification, near-infrared spectroscopy, image dataset, machine learning, cement production, waste management Citation @article{TwinSpectraPaper2026, title = {TwinSpectra - Spectroscopic scans of refuse-derived fuel particles}, author = {Fischer, Jonas and Bikhit, Mina and Wrze\'sniowski, \L{}ukasz and Niepala, Adriana and Mielcarek, Jakub and Illana-Mahiques, Enric and Ruszczak, Bogdan}, journal = {Data in Brief}, year = {2026}, volume = {67}, pages = {112920}, doi = {10.1016/j.dib.2026.112920}, url = {https://www.sciencedirect.com/science/article/pii/S2352340926004713}, issn = {2352-3409}} Instructions and Availability The TwinSpectra dataset, along with code and trained baseline models, is released to ensure full reproducibility of research and machine learning models development. Availability: TwinSpectra Dataset is published on Zenodo. Code: The accompanying package includes Python code (Jupyter Notebooks) for displaying data, reproducing baseline results with the dataset. Acknowledgements We would like to acknowledge the contribution of QZ Solutions, who kindly provided the radiospectrophotometer and calibrated lights that were utilized in the preparation of this dataset. The dataset files measurements_2151.csv- 718 spectroscopic measurements with particle group labels and subset folds with readouts in the range from 350 to 2,500 nm (2151 data features), measurements_125.csv- 720 spectroscopic measurements with particle group labels and subset folds, and registered particle reflectance in the range of 908.1 to 1,676.2 nm (125 data features), data_preview.ipynb - a brief preview of the delivered datasets, baseline_model.ipynb - an exemplary file with a linear model that was trained using this dataset. The only requirements for running these examples are the Python=3.10 with pandas, matplotlib, and SciKit-learn installed.