[A hyperspectral subpixel target detection method based on inverse least squares method].

Area of Science:

• Remote Sensing
• Hyperspectral Imaging
• Signal Processing

Context:

• Subpixel target detection in hyperspectral imagery is crucial for identifying small objects.
• Traditional methods like Orthogonal Subspace Projection (OSP) have limitations in accuracy and computational efficiency.
• The complexity of background spectra often hinders effective target detection.

Purpose:

• To develop and evaluate a novel Inverse Least Squares (ILS) method combined with Mahalanobis distance for subpixel target detection.
• To preprocess hyperspectral data using the SNV algorithm and calculate regression coefficients via Partial Least Squares (PLS).
• To compare the performance of the proposed ILS method against the traditional OSP method using AVIRIS data.

Summary:

• The Inverse Least Squares (ILS) method establishes an inverse model using a target spectrum and pixel spectra, with SNV preprocessing and PLS for regression coefficients.
• Mahalanobis distance is calculated for each pixel's regression coefficient; pixels exceeding a 3-sigma threshold are identified as subpixel targets.
• This approach requires only the target spectrum and is insensitive to background complexity, unlike methods needing background modeling.

Impact:

• The ILS method significantly outperforms the OSP method in subpixel target detection accuracy on AVIRIS remote sensing data.
• The proposed method offers higher detection accuracy and reduced computational time, as evidenced by ROC curves and SNR analysis.
• This advancement provides a more robust and efficient solution for identifying small targets in complex hyperspectral scenes.