Light consistency correction for the liquid crystal tunable filter hyperspectral imaging system.

Area of Science:

• Spectroscopy and Imaging
• Textile Science and Engineering
• Computer Vision and Machine Learning

Background:

• Hyperspectral imaging systems (HIS) provide rich spectral information per pixel for advanced fabric color measurement.
• Traditional colorimeters offer limited spectral data compared to HIS.
• Liquid crystal tunable filter HIS face illumination uniformity and light source consistency issues in fabric color analysis.

Purpose of the Study:

• To develop a robust calibration model for hyperspectral imaging systems to address illumination uniformity challenges.
• To enhance the consistency of fabric color measurements across different light sources.
• To improve the accuracy and reliability of fabric color data obtained via HIS.

Main Methods:

• Developed a uniformity calibration model by integrating gradient boosted decision trees with a sliding window algorithm to correct illumination inconsistencies.
• Employed a deep neural network (DNN) model to correct reflectance measurements under various light sources.
• Merged the uniformity calibration and DNN models into a comprehensive light-consistency correction model.

Main Results:

• The developed light-consistency correction model significantly reduced color differences in fabric samples.
• Average color difference decreased from 3.636 to 0.854, representing a 76.51% enhancement.
• Achieved consistent fabric color measurements despite nonuniform lighting and diverse light sources.

Conclusions:

• The proposed machine learning-based calibration model effectively resolves illumination uniformity and light source consistency issues in HIS fabric color measurement.
• The enhanced accuracy and consistency enable reliable fabric color analysis in real-world, variable lighting conditions.
• This approach advances the application of hyperspectral imaging in textile quality control and color management.