Automated Inorganic Pigment Classification in Plastic Material Using Terahertz Spectroscopy.

Area of Science:

• Materials Science
• Spectroscopy
• Machine Learning

Background:

• Accurate identification of inorganic pigments in plastics is crucial for quality control and material characterization.
• Terahertz (THz) spectroscopy offers a non-destructive method for material analysis, but challenges exist in classifying the resulting 1D data.
• Existing classification algorithms struggle with the complexity and dimensionality of raw 1D THz spectral data.

Purpose of the Study:

• To develop an automated method for classifying inorganic pigments in plastic materials.
• To overcome the limitations of direct 1D terahertz spectral data classification.
• To enhance the efficiency and accuracy of plastic material identification using machine learning.

Main Methods:

• Utilized terahertz spectroscopy to acquire spectral data from plastic materials.
• Developed a novel pre-processing technique to transform 1D THz data into 2D representations (images).
• Employed convolutional neural networks (CNNs) for the classification of the pre-processed 2D THz data.

Main Results:

• The proposed 1D to 2D data transformation significantly improved classification performance.
• The CNN classifier achieved high accuracy in identifying inorganic pigments.
• The method demonstrated robustness against measurement bias and reduced dataset complexity.

Conclusions:

• The novel pre-processing technique combined with CNNs provides an effective solution for automatic plastic material classification.
• This approach outperforms traditional 1D data classification methods like Support Vector Machines and Naive Bayes.
• The developed method offers a promising tool for rapid and accurate material identification in industrial applications.