Dynamical Spectral Unmixing of Multitemporal Hyperspectral Images.

Area of Science:

• Remote Sensing
• Image Analysis
• Signal Processing

Background:

• Hyperspectral imaging captures detailed spectral information, but analyzing time-series data presents challenges.
• Spectral unmixing aims to identify constituent materials and their proportions within a mixed pixel.
• Existing methods often struggle with the dynamic nature of materials over time.

Purpose of the Study:

• To develop a dynamical model for unmixing time-series hyperspectral images.
• To propose an efficient algorithm for estimating latent spectral signatures and fractional abundances.
• To validate the model's performance on both synthetic and real-world data.

Main Methods:

• A dynamical model based on linear mixing processes is formulated for each time instant.
• Spectral signatures and fractional abundances are treated as latent variables with a general dynamical structure.
• An efficient spectral unmixing algorithm is derived using alternating minimization on a simplified model.

Main Results:

• The proposed algorithm effectively estimates latent variables (spectral signatures and abundances) in multitemporal hyperspectral data.
• Demonstrated performance on synthetic datasets validates the model's accuracy.
• Successful application to real multitemporal hyperspectral images showcases practical utility.

Conclusions:

• The dynamical model provides a robust framework for hyperspectral image unmixing over time.
• The derived algorithm offers an efficient and effective solution for analyzing dynamic spectral scenes.
• This approach advances the capabilities of hyperspectral data analysis in various applications.