A Multi-Radical Collaborative Self-Adaptive Denoising Method Based on Res-U-Net and Its Application in CH-PLIF Image Denoising

Planar laser-induced fluorescence (PLIF) is a widely applied non-invasive diagnostic technique in the combustion of power units. PLIF often detects CH and OH radicals to be the key markers for chemical process. Therefore, the clarity of OH-PLIF and CH-PLIF imaging is crucial for studying combustion phenomena. Given the short lifetime of CH radicals in flames, the signal-to-noise ratio (SNR) of CH-PLIF images is much lower than that of OH-PLIF images. Faced with the challenge that traditional denoising methods are ineffective for CH-PLIF images, this study proposes an improved Res-U-Net model based on the U-Net network and incorporating residual connections. This model builds on the traditional U-Net architecture, incorporating previous usage experience and introducing a shortcut connection structure to enhance performance. To improve the model's denoising performance under extreme SNR conditions, this study employs a CH-OH collaborative denoising strategy, which uses OH-PLIF images with a higher SNR to assist in the denoising of CH-PLIF images. Comparisons with several other models on low-SNR datasets demonstrate that the proposed denoising model achieves the best denoising performance.