REDETR-RISTD: Real-Time Long-Range Infrared Small Target Detection Network Based on the Reparameterized Efficient...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Infrared Imaging

Background:

• Detecting small infrared targets at long ranges is challenging due to small size, low signal-to-noise ratio (SNR), and complex backgrounds.
• Existing methods often compromise accuracy for efficiency or vice-versa.

Purpose of the Study:

• To propose a novel real-time long-range infrared small target detection network (REDETR-RISTD) that maintains accuracy while reducing computational complexity.
• To enhance the detection of small targets in infrared imagery.

Main Methods:

• Developed a reparameterized multi-scale feature extraction module (RMSFE) for a lightweight backbone (RepEMSNet).
• Designed an Attention-Based Intra-Scale Contextual Features Interaction (AICFI) module for enhanced feature focus and interaction.
• Implemented a multi-scale pyramid feature fusion network (MSPFN) with bidirectional fusion for improved feature capture.

Main Results:

• REDETR-RISTD achieved competitive performance on public datasets with significantly reduced parameters (13.814 M).
• Demonstrated high AP50 and recall rates across multiple datasets, e.g., 96.5% and 92.7% on one.
• Successfully balanced detection accuracy and computational efficiency.

Conclusions:

• REDETR-RISTD offers an effective solution for real-time long-range infrared small target detection.
• The proposed modules (RMSFE, AICFI, MSPFN) contribute to improved accuracy and efficiency.
• This network provides a viable trade-off between detection performance and computational cost.