Deep Learning-based Visible and Infrared Image Fusion using Dual-Branch Attention Network for Enhanced Structural Clarity

This paper presents a deep learning-based framework for the fusion of visible and infrared images to enhance image visibility and structural clarity under varying conditions. The proposed model employs a dual-branch encoder to independently extract features from each modality, followed by a channel attention mechanism to adaptively weight and combine these features. A decoder reconstructs the fused image while preserving essential details from both inputs. The model is trained end-to-end using a composite loss function that emphasizes pixel accuracy, structural similarity, and edge consistency. Experimental evaluations demonstrate that the proposed method achieves superior performance across multiple objective metrics, including MS-SSIM (0.9250), Mutual Information (7.8500), and Entropy (6.1200). Compared to existing techniques such as DTCWT, LatLRR, and NSCT-SR, the proposed model shows significant improvements in structural and perceptual quality. The results confirm the model’s robustness and effectiveness, making it suitable for real-world applications like night vision, surveillance, and autonomous navigation.