Systematic reduction of hyperspectral images for high-throughput plastic characterization

Area of Science:

• Chemometrics
• Spectroscopy
• Data Science

Background:

• Hyperspectral imaging (HSI) offers detailed chemical analysis but faces challenges with large datasets and redundancy.
• Efficient data analysis is critical for HSI applications, particularly in resource-limited areas like waste sorting.
• Non-Negative Matrix Factorization (NMF) is a key algorithm for HSI chemical resolution, benefiting from data reduction.

Purpose of the Study:

• To enhance the speed and efficiency of hyperspectral imaging data analysis.
• To develop a data reduction method for improving Non-Negative Matrix Factorization (NMF) performance in HSI.
• To enable more straightforward data investigation and analysis for applications like plastic sorting.

Main Methods:

• Application of a convex-hull method for pixel and wavelength selection.
• Removal of uninformative and redundant information from HSI datasets.
• Utilizing chemometric approaches for automated and evidence-based data reduction.

Main Results:

• Significant reduction in HSI data size and computational strain.
• Effective elimination of highly mixed pixels and data redundancy.
• Demonstrated improved data analysis quality and efficiency on simulated and real HSI data for plastic sorting.

Conclusions:

• The convex-hull method provides an effective strategy for HSI data reduction.
• This approach enhances the performance of NMF for chemical resolution and waste sorting.
• Streamlined HSI data analysis facilitates more straightforward investigation and application in waste management.