Firefighting Infrared-and-Visible Image Fusion via Re-Parameterized Laplacian and Attention-Based Network

In urgent firefighting scenarios, extracting and fusing features from different images rapidly is crucial for providing comprehensive environmental information in time. Here we propose a Re-parameterized Laplacian and Attention-based Image Fusion Network (RLAFusion), which combines infrared and visible light images to enhance visual information display. The Re-parameterized Laplacian (RL) encoder extracts features from infrared and visible images. The extracted feature maps are subsequently refined by a Deconvolution and Self-Attention (DSA) mechanism-based decoder. RLAFusion outperforms other image fusion approaches in terms of balancing inference speed with image quality, including aspects such as image contrast, image perceptual quality, signal fidelity, and information richness. Moreover, we find that the re-parameterization process could increase inference speed by nearly 40%. This enhances firefighting vision, improving the detection of fire sources and hazards, which could enhance automated decision-making in firefighting robots.