Multi-Scale Visible and Infrared Fusion Target Detection Method Based on a Dual-Branch Architecture

In target detection, efficient fusion of visible and infrared images is crucial. Existing CNNs, which rely on local information extraction, struggle to capture global information, affecting fused image quality. Additionally, fusion networks often lose high-frequency information, reducing target detection accuracy. To address these issues, this paper presents a dual-branch architecture combining a lightweight Transformer with a CNN incorporating CBAM attention mechanisms. The branches handle low-frequency global features and high-frequency local information, improving fused image quality and target detection accuracy. Experiments on MSRS and LLVIP datasets show the proposed algorithm outperforms state-of-the-art methods in fusion performance. Quantitative analysis demonstrates significant improvements in metrics like Information Entropy (EN), Spatial Frequency (SF), and Average Gradient (AG), proving its superiority in detail and frequency representation.