Near-infrared hyperspectral circular polarization imaging and object classification with machine learning.

Area of Science:

• Optics and Photonics
• Machine Learning
• Materials Science

Background:

• Conventional near-infrared (NIR) hyperspectral imaging faces challenges in classifying materials with similar absorption spectra.
• Object classification is crucial for applications like waste management and recycling.

Purpose of the Study:

• To develop and evaluate a novel hyperspectral circular polarization (S3) imaging system in the NIR region.
• To assess the performance of hyperspectral S3 imaging for material classification compared to conventional methods.
• To explore the potential applications of this technology in waste classification.

Main Methods:

• Construction of a hyperspectral S3 imaging system using a circularly polarized broadband light source, a polarization grating, and a hyperspectral camera.
• Acquisition of hyperspectral S3 images of various plastic samples.
• Application of machine learning algorithms for classification based on the captured hyperspectral S3 data.

Main Results:

• The hyperspectral S3 imaging system successfully captured detailed spectral and polarization information.
• Machine learning classification using hyperspectral S3 images achieved higher precision than conventional NIR hyperspectral images.
• The system demonstrated effectiveness in distinguishing between plastic samples with similar absorption spectra.

Conclusions:

• Hyperspectral S3 imaging offers enhanced capabilities for object classification, particularly for materials with subtle spectral differences.
• This technology holds significant potential for improving the accuracy and efficiency of garbage classification in recycling facilities.
• The developed system provides a valuable tool for advanced material analysis and identification.