LRAF-Net: Long-Range Attention Fusion Network for Visible-Infrared Object Detection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Visible-infrared object detection leverages complementary image data.
• Existing methods often neglect long-range dependencies between modalities, limiting performance in complex environments.

Purpose of the Study:

• To propose a feature-enhanced long-range attention fusion network (LRAF-Net) for improved visible-infrared object detection.
• To address limitations in current methods by effectively fusing long-range dependencies between enhanced visible and infrared features.

Main Methods:

• Utilized a two-stream CSPDarknet53 for feature extraction.
• Introduced a novel data augmentation method with asymmetric complementary masks.
• Developed a cross-feature enhancement (CFE) module and a long-range dependence fusion (LDF) module incorporating positional encoding.

Main Results:

• The LRAF-Net effectively fuses long-range dependencies from enhanced visible and infrared features.
• Experiments on VEDAI, FLIR, and LLVIP datasets demonstrated superior performance compared to existing methods.
• Achieved state-of-the-art results in visible-infrared object detection.

Conclusions:

• The proposed LRAF-Net significantly enhances object detection performance by effectively fusing complementary visible and infrared image information.
• The integration of long-range attention and feature enhancement modules is crucial for handling complex scenes.
• The method shows strong potential for real-world applications requiring robust object detection across different imaging conditions.