Hyperspectral imaging dataset. Calibration, validation and test datasets for the classification of recyclable materials.

Hyperspectral spectral images of 5 types of recyclable materials: cardboard, HDPE, PS, PET and PP. Spectral imaging was carried out in three stages, generating three independent datasets for calibration, validation and testing of the results. For the calibration dataset, one object of each one of the 5 recyclable materials was imaged separately. For the validation dataset, a single mixed image, presenting a mixture of the 5 types of recyclable materials was acquired. Each type of material was cut in a different shape to facilitate the visual interpretation of the results. The same procedure was independently repeated for the test dataset. Spectral images of the samples, in the NIR range (943 - 1643 nm), were acquired using a pushbroom hyperspectral imaging system (Spectral Scanner, DV Optics Ltd, Padova, Italy). The Spectral resolution of the system was 7nm and the spatial resolution of the setup was approximately 170 × 170 μm per pixel. Samples were placed on a white tile for image acquisition. If you use this dataset please reference the published article as: Herrero-Langreo, A., Altun, N., & Gowen, A. A. A. (2026). Optimizing Distance‐Based Classification in Hyperspectral Imaging: a Tutorial on the Influence of Spectral Pretreatments. Journal of Chemometrics, 40(2), e70101. https://doi.org/https://doi.org/10.1002/cem.7010 Table 1. Dataset description, number of pixels and objects per dataset. Material Dataset 1: Calibration, 1 image per material. (N of pixels/N objects) Dataset 2: Validation, one single validation image with mixed materials (N of pixels/N objects) Dataset 3: Test, one single validation image with mixed materials (N of pixels/N objects) Carboard 20 810 pixels /1 object 930 pixels / 9 objects 970 pixels/8 objects HDPE 24 675 pixels /1 object 3 789 pixels / 6 objects 4039 pixels/5 objects PS 7 867 pixels /1 object 1 952 pixels / 5 objects 1866 pixels/8 objects PET 21 422 pixels /1 object 2 250 pixels / 5 objects 4841 pixels /8 objects PP 12 034 pixels /1 object 4 655 pixels / 5 objects -