Enhanced synthetic aperture radar automatic target recognition method based on novel features.

Area of Science:

• Computer Vision
• Signal Processing
• Machine Learning

Background:

• Automatic target recognition (ATR) in synthetic aperture radar (SAR) images is crucial for intelligence, surveillance, and reconnaissance.
• Existing feature representations often struggle with aspect variations and capturing comprehensive target information.

Purpose of the Study:

• To propose novel, rich feature representations for SAR ATR.
• To enhance discrimination and robustness against aspect variations.
• To integrate these features into a unified recognition framework.

Main Methods:

• Feature extraction using monogenic signal and polar mapping to capture spatial-spectral properties.
• Development of a new target-shadow feature for geometric information.
• Integration of proposed features within a multiple kernel learning framework.

Main Results:

• Demonstrated effectiveness of monogenic signal for SAR target recognition.
• Polar mapping successfully reduced aspect variation influence.
• The unified framework achieved high recognition accuracy on the MSTAR dataset.

Conclusions:

• The proposed feature representations offer significant improvements for SAR ATR.
• The unified multiple kernel learning framework provides a robust and accurate solution.
• This work advances the state-of-the-art in SAR image-based target recognition.