Deep learning-based suppression of cold reflections in infrared systems

This study proposes a deep learning-based method to suppress cold reflection artifacts in cooled mid-wave infrared imaging systems. First, a mid-wave infrared optical system is designed, followed by a cold reflection simulation using the Narcissus macro. Second, an equivalent temperature difference superposition method is adopted to generate synthetic infrared images with cold reflection patterns, thereby establishing a multi-scene dataset. Furthermore, an enhanced NU-Net architecture incorporating L1-perceptual hybrid loss optimization is developed for cold reflection artifact suppression. Experimental results demonstrate that the proposed framework significantly improves the reconstructed image quality, achieving superior PSNR and SSIM metrics compared to conventional methods. This methodology realizes high-quality infrared imaging with minimized cold reflection interference by virtue of a relatively compact optical system.