Multispectral Household Plastic Material Database (MHPM)

Effective plastic recycling requires the accurate identification of polymer types. However, existing optical sorting systems often struggle to distinguish common household plastics reliably. To overcome this identification bottleneck, multispectral imaging is necessary to capture the unique near-infrared spectral fingerprints of different materials. To address this, a comprehensive multispectral household plastic material database (MHPM) was developed, which can be processed to train and evaluate automated material classifiers. Specifically, the database contains multispectral recordings of the seven most common household plastics (PET, PP, PS, HDPE, LDPE, PVC, PTFE), captured by a nine-camera array setup equipped with distinct NIR bandpass filters. The approach to extract the spectral fingerprints involves a dedicated preprocessing pipeline that spatially aligns and normalizes the raw images to derive features such as wavelength intensity differences and false-color representations. Of course, this database can also be used for many other tasks, where reliable material identification is necessary, for example, evaluating machine learning models, optimizing spectral band selection, and developing cost-efficient industrial sorting pipelines.

Database

The actual data can be found here: https://github.com/FAU-LMS/MHPM The MHPM database provides 25 scenes recorded from 800 nm to 1600 nm in 25 nm and 50 nm steps, resulting in 24 multispectral channels. Furthermore, one RGB image of each scene is contained. Additionally, the Spectral fingerprints of each material class are included. The images are cropped/upsampled to a resolution of 1740 x 1348 pixels. The images are already calibrated, normalized, and registered and reconstructed.

Camera specifics

Grayscale cameraRGB cameraShort-wave infrared camera (SWIR)
Camera typeAllied Vision Alvium 1800 U-1240mAllied Vision Alvium 1800 U-1240cAllied Vision Alvium 1800 U-130m VSWIR
Focal length5mm5mm60.1mm
Image sensorSony IMX226Sony IMX226Sony IMX990
Sensor pixel size1.85 µm
1.85 µm

5 µm
Resolution4024 x 30364024 x 30361296 x 1032
Frame rate35 fps35 fps130 fps
Bit depth101010
Lens transmission200 nm – 1150 nm300 nm – 1100 nm400 nm – 1700 nm

Publication

If you use the dataset or source code for your research, you should cite the following paper:

Katja Kossira, Jürgen Seiler, André Kaup
Multispectral household plastic classification for recycling using a camera array
EURASIP Journal on Advances in Signal Processing, vol. 2026, num. 66, 2026, pp. 1-25

DOI: SpringerNature
arxiv: arXiv

License

The database and source are licensed under the Creative Commons CC BY license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Contact

lms-datasets@fau.de