Artificial Intelligence for Pigment Classification Task in the Short-Wave Infrared Range.

Area of Science:

• Materials Science
• Art Conservation
• Spectroscopy

Background:

• Hyperspectral reflectance imaging in the short-wave infrared (SWIR) range (1000-2500 nm) enhances paint material characterization.
• Interpreting complex SWIR spectra with overlapping features poses a significant challenge for detailed analysis.
• High dimensionality and spectral complexity require advanced data treatment methods for accurate results.

Purpose of the Study:

• To develop and test an innovative data treatment approach for SWIR hyperspectral imaging of paint materials.
• To combine a pigment spectral database with a deep neural network (DNN) for enhanced pigment classification.
• To address the challenges of interpreting complex SWIR spectral data in cultural heritage applications.

Main Methods:

• Utilized hyperspectral reflectance imaging in the SWIR range (1000-2500 nm).
• Developed a pigment spectral database from model and historical paint samples.
• Applied a deep neural network (DNN) for multi-label pigment classification at the pixel level.

Main Results:

• The combined pigment spectral database and DNN approach enabled effective multi-label pigment classification within SWIR hyperspectral data cubes.
• Demonstrated successful application on both pigment reference samples and a historical Buddhist painting (thangka).
• Compared and discussed results against conventional Spectral Angle Mapping techniques.

Conclusions:

• The DNN-based approach offers a robust solution for analyzing complex SWIR hyperspectral data of paint materials.
• This method significantly improves pigment identification and characterization in cultural heritage objects.
• The integration of spectral databases and DNNs represents a promising advancement in hyperspectral imaging analysis.