[Three-dimensional hybrid denoising algorithm in derivative domain for hyperspectral remote sensing imagery].

Area of Science:

• Remote Sensing
• Image Processing
• Signal Processing

Context:

• Hyperspectral remote sensing imagery is crucial for environmental monitoring and resource management.
• Noise in hyperspectral data degrades image quality and hinders accurate analysis.
• Existing denoising methods often struggle with the complex noise characteristics of hyperspectral imagery.

Purpose:

• To propose a novel three-dimensional hybrid denoising algorithm for hyperspectral remote sensing imagery.
• To enhance the signal-to-noise ratio (SNR) of hyperspectral data.
• To address the limitations of existing denoising techniques in the spectral derivative domain.

Summary:

• The proposed algorithm transforms hyperspectral imagery into the spectral derivative domain to amplify noise.
• A wavelet-based Bayes-Shrink method is applied in the spatial domain, while the Savitzky-Golay filter smooths the spectral domain considering band-wise noise variance.
• The processed data is integrated spectrally and corrected for errors, demonstrating efficient noise reduction and significant SNR improvement.

Impact:

• The algorithm effectively reduces noise in hyperspectral imagery, improving data quality for analysis.
• Achieves a substantial SNR improvement from 600:1 to over 2000:1, as validated on AVIRIS data.
• Provides a robust solution for denoising hyperspectral remote sensing data, enabling more reliable scientific interpretations.