Interactive Visualization of Hyperspectral Images Based on Neural Networks.

Area of Science:

• Data visualization
• Machine learning
• Spectroscopy

Background:

• Hyperspectral images present interpretation challenges due to their high dimensionality (hundreds of bands).
• Existing hyperspectral image analysis techniques often lack intuitive visualization and user interaction.
• Identifying distinct substances within hyperspectral data can be difficult with current methods.

Purpose of the Study:

• To develop an intuitive visualization tool for hyperspectral images using neural networks.
• To enable users to interactively select regions and bands of interest for analysis.
• To facilitate the identification and classification of substances within hyperspectral data.

Main Methods:

• A neural network-based visualization tool was developed for hyperspectral images.
• A cascade neural network was trained to generate scatterplots from hyperspectral features.
• User-defined cluster centers were used to train the neural network for accurate classification.

Main Results:

• The tool generates scatterplots that visually represent clusters of hyperspectral features.
• The scatterplots reveal relationships between different substances within the data.
• The trained neural network can be reused for time-varying hyperspectral data analysis without retraining.

Conclusions:

• The developed visualization solution provides an intuitive way to analyze hyperspectral images.
• Domain experts can remain involved in the analytical loop, enhancing interpretation.
• The tool effectively identifies different substances, overcoming limitations of existing techniques.