Continuum removal for ground-based LWIR hyperspectral infrared imagery applying non-negative matrix factorization

Area of Science:

• Remote Sensing
• Spectroscopy
• Image Analysis

Background:

• Continuum removal is crucial for hyperspectral image analysis, but often relies on approximations for blackbody temperature calculations.
• Existing methods for hyperspectral data analysis can be limited by assumptions in background spectral calculations.

Purpose of the Study:

• To introduce a novel method for continuum removal in hyperspectral infrared imagery.
• To eliminate the need for calculating background blackbody temperature spectra in ground-based applications.

Main Methods:

• A laboratory experiment using long-wave infrared (LWIR) imaging with a heating source and two acquisition sets.
• Application of Non-negative Matrix Factorization (NMF) with Gradient-Descent (GD) and Non-Negative Least-Squares (NNLS) optimization algorithms.
• Estimation of downwelling radiance and comparison with conventional methods using spectral angle mapper and normalized cross-correlation (NCC) with additive noise.

Main Results:

• NMF-GD achieved an average similarity of 72.5% using NCC.
• NMF-NNLS demonstrated an average similarity of 77.6% using NCC.
• Both NMF-GD and NMF-NNLS showed robust performance under varying levels of additive noise (1%-20%).

Conclusions:

• The proposed NMF-based continuum removal method is effective for LWIR hyperspectral imagery.
• This approach offers a more accurate and assumption-free alternative for hyperspectral data analysis.
• NMF-NNLS provides superior performance compared to NMF-GD and conventional methods.