Robust Small Target Co-Detection from Airborne Infrared Image Sequences.

Area of Science:

• Computer Vision
• Signal Processing
• Infrared Imaging

Background:

• Detecting small targets in infrared images with complex backgrounds is challenging due to background clutter and target-like interference.
• Existing methods often struggle with accurately distinguishing small targets from intricate background features.

Purpose of the Study:

• To propose a novel infrared target co-detection model that leverages both background self-correlation and target commonality in the spatio-temporal domain.
• To enhance the detection accuracy and robustness of small targets in infrared image sequences.

Main Methods:

• A dense target extraction model using nonlinear weights to suppress backgrounds and enhance small targets.
• A sparse target extraction model employing entry-wise weighted robust principal component analysis with local weighted entropy for accurate target extraction and clutter suppression.
• A target refinement model utilizing spatio-temporal commonality for false alarm suppression and target confirmation, including tracklet association.

Main Results:

• The proposed model effectively suppresses background clutters in infrared images.
• Accurate detection of small targets is achieved, even in the presence of target-like interference.
• The co-detection method demonstrates superior performance in discriminating small targets from background noise.

Conclusions:

• The novel co-detection model provides an effective solution for small target detection in complex infrared backgrounds.
• The integration of dense and sparse extraction with spatio-temporal analysis significantly improves detection accuracy and reduces false alarms.
• This approach offers a robust method for identifying real targets and confirming their presence through trajectory analysis.