Long-Wave and Short-Wave Infrared Image Conversion Based on Diffusion Model

Infrared images play a significant role in various applications, including ship detection, as they can display information about target objects in diverse complex environments. However, owing to the different formation mechanisms of long-wave and short-wave images in infrared imaging, the information they hold varies. In this paper, we propose a long-wave and short-wave infrared image conversion method based on a diffusion model to fully utilize the information in both long-wave and short-wave images, avoiding the issue of insufficient information when only a single long-wave or short-wave image can be obtained in certain situations. Specifically, we first propose a collaborative training strategy based on the diffusion model. By minimizing the optimal transport distance between the source and target images, the edge and structural information are preserved while smoothing the image regions. Next, we combine attention with traditional optimization algorithms, allowing the model to better focus on essential areas in the input image and achieve a smooth transition between the two image domains. Finally, we adopt the optimal transport algorithm based on the Sinkhorn to accelerate the model's execution speed. Simulation results demonstrate that the proposed method effectively improves training efficiency, outperforming baseline image generation methods in terms of SSIM and PSNR values on the selected dataset, while reducing the LPIPS value. The proposed method has promising application potential in the field of ship detection.