Infrared Small and Dim Target Detection With Transformer Under Complex Backgrounds.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Infrared Imaging

Background:

• Infrared small and dim (S&D) target detection is critical for infrared search and tracking systems.
• Existing deep learning methods struggle with detecting S&D targets due to limited range dependency capture by convolutional neural networks.
• The inherent characteristics of S&D targets increase the risk of missed detections.

Purpose of the Study:

• To develop a robust and generalizable infrared S&D target detection method.
• To overcome the limitations of existing deep learning approaches in capturing long-range dependencies.
• To reduce missed detections of small and dim infrared targets.

Main Methods:

• Utilized a transformer architecture with a self-attention mechanism to learn long-range feature correlations.
• Designed a feature enhancement module to learn discriminative features for S&D targets.
• Employed a U-Net-like decoder with skip connections to preserve target information and a segmentation head for final detection.

Main Results:

• The proposed transformer-based method demonstrated superior performance compared to state-of-the-art methods on two public datasets.
• Achieved enhanced detection accuracy for small and dim infrared targets.
• Showcased improved generalization ability and noise tolerance.

Conclusions:

• The transformer-based approach effectively addresses the challenges in infrared S&D target detection.
• The method offers a robust solution with improved accuracy and generalization capabilities.
• This work advances the field of infrared search and tracking systems.