Band selection method for subpixel target detection using only the target reflectance signature.

Area of Science:

• Remote Sensing
• Image Processing
• Data Science

Background:

• High dimensionality of hyperspectral images poses challenges for information extraction.
• Dimension reduction techniques, like band selection, are crucial for efficient data processing.
• Existing band selection methods for subpixel target detection often require background spectral signatures, limiting their applicability.

Purpose of the Study:

• To develop a novel band selection method for subpixel target detection that utilizes only target spectral reflectance signatures.
• To overcome the limitation of requiring background information in traditional band selection methods.
• To enable efficient target detection in scenarios where background data is unavailable.

Main Methods:

• Developed a band selection method based solely on target spectral reflectance signatures.
• Validated the method using a utility prediction model and real hyperspectral images.
• Cross-validated results with simulated images containing perfect ground truth data.
• Compared detection statistics with three other established band selection methods.

Main Results:

• The proposed method demonstrated effective band selection for subpixel target detection without background information.
• Validation and cross-validation confirmed the method's performance on real and simulated data.
• Detection statistics showed comparable or improved results against other band selection techniques.

Conclusions:

• The developed method offers a viable solution for subpixel target detection when background information is absent.
• This approach can optimize adaptive spectral imaging systems, reducing data handling and processing costs.
• The method's simplicity allows for on-board computation and adaptable band selection criteria for changing targets.