FLFusionSR: a fast and lightweight fusion and super-resolution network for infrared and visible images on edge devices

In real-world applications, the application of image fusion and super-resolution networks is limited by the computational burden and power consumption of edge devices. To resolve this problem, we propose a fast and lightweight network for infrared and visible image fusion and super-resolution called FLFusionSR. It directly learns end-to-end mapping from the original low-resolution infrared and visible images to a single high-resolution fused image. To meet the hardware requirements and real-time processing needs of edge devices, we adopt lightweight strategies in network design. To improve network performance without increasing computational complexity, we employ an auxiliary learning strategy during network training. The running analysis shows that our proposed method achieves 30FPS running speed on an edge device with 7.5 W power consumption, meeting the real-time processing requirements. Quantitative and visual comparison experiments on two datasets demonstrate that our proposed algorithm achieves state-of-the-art performance. The code is available at https://github.com/bearxwm/FLFusionSR.