Methods

Differential Scanning Calorimetry (DSC)
The differential scanning calorimetry (DSC) of the samples was performed with a PerkinElmer DSC 8500 (USA). Samples of 4–10 mg (Sartorius, MSE125P-000-DU) were sealed in aluminum crucibles (50 µL) with pierced lids (Pans and covers type: 02190041). The heat-cool cycle was performed twice followed by a StepScan. All scans were carried out under a nitrogen atmosphere with a flow rate of 20 mL min^-1. The samples were heated from 20.00 to 280.00 °C at 10.00 °C min⁻¹ and held for 2.0 min, before being cooled down from 280.00 (depending on polymer) to 20.00 °C at 60.00 °C min⁻¹ and held for 2.0 min. The samples were then reheated from 20.00 to 280.00 °C at 10.00 °C min⁻¹, before being cooled down from 280.00 to 20.00 °C at 60.00 °C min⁻¹. Finally, StepScan with step size 1.0 °C was performed from 20.0 to 280.0 °C at 5.0 °C min^-1. The sample temperature is plotted against the Heat Flow [mW].

Thermogravimetric Analysis (TGA)  
Samples of 5−10 mg (Sartorius, MSE125P-000-DU) are heated in laboratory air (TGA AIR) and under nitrogen (TGA N2) in a TG 209 F1 Libra (Netzsch, GE) from 35 °C to 800 °C, 10 °C min−1. The percentage of mass is then plotted against the temperature.

Hyperspectral Imaging (Vis and SWIR)
The dual-camera hyperspectral imaging system (Newtec Engineering A/S, Denmark) consists of hyperspectral cameras positioned 55 cm above a conveyor belt (29 cm wide, speed 62.5 mm s^-1) transporting the samples. The illumination of the conveyor belt was done by two units of four halogen lamps (12V, 20W) placed 30 cm above the conveyor belt at a 45° angle. The cameras used are a) a line-scan hyperspectral camera (QT5022, Qtechnology A/S, Denmark) with an in-build spectrograph, a VS-H1618-IRC/11 (VS Technology, Japan) objective lens, and a GSense2020 (GSense, Gpixel, Belgium) metal-oxide-semiconductor CMOS sensor for the Vis/NIR region, and b) a line-scan hyperspectral camera (QT5022, Qtechnology A/S, Denmark) with an in-build spectrograph, a LM16HC-SW (Kowa, Japan) objective lens, and an InGaAs-Matrix 320×256-C sensor (InGaAs, Andanta, Germany) for the SWIR region. The system was controlled by the onboard software (Newtec Engineering A/S, Denmark). Prior to measurement, a full spatial, spectral, and intensity calibration was performed as previously described. [A] In short; a measurement on a calibration board containing a chess-pattern, LEDs, and TiO₂ powder gives the calibration parameters. The spatial resolution is 4.44 pix mm^-1 (across) by 2.02 pix mm^-1 (along) for Vis/NIR and 0.90 pix mm^-1 (across) by 2.01 pix mm^-1 (along) for SWIR. The spectral resolution is 1.75 nm pix^-1 for Vis/NIR and 8.82 nm pix^-1 for SWIR. Intensity calibration was referenced to TiO₂.

[A] “One step calibration of industrial hyperspectral cameras”
https://doi.org/10.1016/j.chemolab.2022.104609

Hyperspectral Imaging (Vis-SWIR)
The single-camera hyperspectral imaging system (Newtec Engineering A/S, Denmark) consists of hyperspectral cameras positioned 48 cm above a conveyor belt (34 cm wide, speed 80 mm s^-1) transporting the samples. Illumination is provided by two sets of six and seven halogen lamps (12 V, 20 W) irradiating the conveyer belt and samples at a 45° angle. The camera used is a line-scan hyperspectral camera (QT5222, Qtechnology A/S, Denmark) with an in-build spectrograph and an IMX990 sensor (IMX990, Sony, Japan) covering the Vis-SWIR spectral regions. The system was controlled by the onboard software (Newtec Engineering A/S, Denmark). Prior to measurement, a full spatial, spectral, and intensity calibration was performed as previously described. [A] In short; a measurement on a calibration board containing a chess-pattern, LEDs, and TiO₂ powder gives the calibration parameters. The spatial resolution is 4.98 pix mm^-1 (across) by 1.31 pix mm^-1 (along). The spectral resolution is 1.43 nm pix^-1 for the Vis-SWIR spectral region. Intensity calibration was referenced to TiO₂.

Attenuated Total Reflectance Fourier Transformed Inferred Spectroscopy (ATR-FTIR)
Sample 1-261:
Fourier-transformed infrared spectra (FTIR) of samples were collected in an attenuated total reflection (ATR) mode on an iS5 spectrophotometer (Thermo Fisher Scientific, USA) fitted with a ZnSe crystal (iD5, Thermo Fisher Scientific, USA). Background (n = 16) and sample measurement (n = 16) were measured with a resolution of 2 cm⁻¹. Wavelength-dependent penetration depth and baseline were corrected with OMNIC (v. 9.2.98., Thermo Scientific, USA) built-in functions.

Sample 262 and ongoing:
Fourier-transformed infrared spectra (FTIR) of samples were collected in an attenuated total reflection (ATR) mode on an IRSpirit-T (Shimadzu Corporation, JP) fitted with a QATR-S module equipped with a prism made of diamond (Shimadzu Corporation, JP). Background (n = 16) and sample measurement (n = 16) were measured with a resolution of 2 cm⁻¹. Wavelength-dependent penetration depth and baseline were corrected with LabSolutions IR (v2.27., Shimadzu Corporation, JP) built-in functions.

Nuclear Magnetic Resonance (NMR)
Bruker Avance III 600 spectrometer equipped with a 5-mm ¹H TXI probe (Bruker BioSpin, Rheinstetten, Germany), ¹H NMR spectra were acquired at 298 K and a ¹H frequency of 600.13 MHz. ¹H-NMR samples are recorded with 16 scans and d-CHCl3 is used as solvent.

Terahertz time-domain spectroscopy (tHz)
The samples were measured with a fiber-coupled, commercial THz-TDS spectrometer manufactured by TOPTICA Photonics (TeraFlash pro). The setup was arranged in a transmission configuration using four off-axis parabolic mirrors between the fiber-coupled THz emitter and receiver (see Fig. 1b in [B]). A 50 mm focal length parabolic mirror collimates the THz radiation from the emitter, while a parabolic mirror with a focal length of 100 mm focuses it onto the sample. Likewise, a 100 mm focal length parabolic mirror collimates the THz radiation transmitted through the sample, and a 50 mm focal length parabolic mirror focuses it into the receiver. For each sample, ten measurements at random positions on the sample were recorded followed by a single reference measurement where the sample was absent. All the measurements were performed under the same ambient experimental conditions recording time traces with a length of 50 ps and 1000 acquisitions (scan speed: 60 traces/s).

[B] “Identification of black plastics with terahertz time-domain spectroscopy and machine learning”
https://doi.org/10.1038/s41598-023-49765-z

BigSMILES
The BigSMILES annotate the polymers used in the database and follow the guidelines provided by [B].
[C] “BigSMILES: A Structurally-Based Line Notation for Describing Macromolecules” https://doi.org/10.1021/acscentsci.9b00476