Data Mining from Near-infrared Spectra of Plastics by 1/f Fluctuation. (2). The Case of Flat Spectra of Dark Colored Plastics.

小野寺 光永, 長嶋 雲兵, 加藤 澄恵, Haruo Hosoya, 後藤 成志, 天野 敏男, 田辺 和俊, 上坂 博亨Journal of Chemical Software 1999 · DOI

Abstract

Dark-colored plastics have flat NIR spectra with very small differences in reflection intensity. However, using Fast Fourier Transform (FFT) for 1/f fluctuation analysis allows for fast and accurate classification of these plastics.

Introduction

Efficient recycling of plastic waste requires rapid and accurate sorting of various polymers, such as Polypropylene (PP) and Polycarbonate (PC). Near-infrared (NIR) reflection spectroscopy is widely used for this purpose due to its speed and non-destructive nature. However, dark-colored plastics, often containing carbon black, pose a significant challenge. These materials absorb most NIR radiation, resulting in low-intensity, "flat" spectra that lack the characteristic absorption peaks typically used for identification.

Methodology: 1/f Fluctuation Analysis

To overcome the lack of distinct spectral features, this research employs 1/f fluctuation analysis. This technique focuses on the statistical "texture" of the signal rather than its absolute shape.

  1. Data Acquisition: NIR reflection spectra are measured in the 1.6 to 2.2 $\mu$m range.
  2. Fast Fourier Transform (FFT): The spectral data, despite appearing flat, contains minute fluctuations that are analyzed in the frequency domain.
  3. Power Spectrum Mapping: The power spectrum is plotted on a log-log scale. A linear relationship, characteristic of 1/f noise, is typically observed.
  4. Feature Extraction: Linear regression is performed on the log-log plot. The resulting slope and y-intercept serve as unique coordinates to map each plastic sample onto a 2D classification plane.

Experimental Datasets

The study evaluated two main groups of samples:

  • Common Black Plastics: Five types prevalent in electronics and automotive industries: High Impact Polystyrene (HIPS), Acrylonitrile Butadiene Styrene (ABS), Acrylonitrile Styrene (AS), Polycarbonate (PC), and Polypropylene (PP).
  • Polypropylene Series: A set of PP samples with varying shades to test the method's sensitivity to pigment concentration and overall darkness.
Near-infrared reflection spectra of five common black plastics (HIPS, ABS, AS, PC, and PP), showing their relatively featureless and "flat" nature.

Results and Discussion

The 1/f fluctuation analysis successfully extracted distinguishing features from the seemingly featureless spectra.

Classification Accuracy

Even when visual inspection of the NIR spectra suggested nearly identical profiles, the slope-intercept mapping clearly separated the five plastic types. For instance, PP and PC occupied distinct regions in the 2D feature space, allowing for unambiguous automated sorting.

Comparison of NIR spectra for two different Polypropylene (PP) samples, illustrating how intensity varies while the underlying spectral "texture" remains.

Robustness to Darkness

In the PP series, while the y-intercept of the regression line shifted with the sample's darkness (reflecting overall reflectivity), the slope remained characteristic of the base polymer. This demonstrates that the "noise" in the flat spectrum carries a robust signature of the molecular structure.

Conclusion

By applying 1/f fluctuation analysis via FFT, hidden information can be extracted from the flat NIR spectra of dark plastics. This method is computationally efficient and provides a practical solution for high-speed industrial sorting in recycling facilities. Future work will focus on the physical interpretation of the 1/f signatures and expanding the library to include complex polymer blends.

Summary updated

Abstract

暗色のプラスチックの近赤外反射スペクトルは、反射強度の違いが極端に小さく、平板なスペクトルしか得られないが、そのような暗色のプラスチックの平板な近赤外反射スペクトルを用いても、高速フーリエ変換FFTによる1/f揺らぎ解析により、高速かつ実用上十分な精度でプラスチックを分類できることがわかった。
BibTeX
@article{1999data,
  author = {小野寺 光永 and 長嶋 雲兵 and 加藤 澄恵 and Haruo Hosoya and 後藤 成志 and 天野 敏男 and 田辺 和俊 and 上坂 博亨},
  title = {{Data Mining from Near-infrared Spectra of Plastics by 1/f Fluctuation. (2). The Case of Flat Spectra of Dark Colored Plastics.}},
  journal = {Journal of Chemical Software},
  year = {1999},
  doi = {10.2477/jchemsoft.5.103},
}
Figure