An Infrared Small Target Detection Method Based on Attention Mechanism.

Area of Science:

• Computer Vision
• Signal Processing
• Artificial Intelligence

Background:

• Human visual attention excels at rapid infrared target recognition and scene adaptability.
• Existing methods struggle with complex backgrounds and small infrared targets.

Purpose of the Study:

• To propose an attention-based infrared small target detection method.
• To improve detection accuracy and reduce false alarms in complex infrared imagery.

Main Methods:

• A three-module approach: bottom-up passive attention, top-down active attention, and decision feedback equalization.
• Top-down module: Gaussian shape feature extraction and quaternion cosine transform for multi-dimensional fusion.
• Bottom-up module: Optimal fast local contrast, improved circular pipeline filtering, and multi-scale Laplacian of Gaussian filter for candidate region identification and size estimation.

Main Results:

• The proposed method demonstrates superior detection performance compared to baseline methods (RLCM, ILCM, PQFT, MPCM, ADMD).
• Effective reduction of false alarms in complex backgrounds like sea, sky, and ground clutter.
• Mathematical proofs validate the method's efficacy.

Conclusions:

• The attention-based method significantly enhances infrared small target detection.
• The integration of multi-dimensional features and robust filtering contributes to improved accuracy and adaptability.
• The method offers a promising solution for real-world infrared surveillance applications.