TDLAS signal denoising based on PCA-GASF-DCResNet

To mitigate the impact of external noise interference in the second-harmonic spectrum during gas concentration measurements using tunable diode laser absorption spectroscopy (TDLAS), we propose a dual-channel convolutional neural network algorithm with multilevel feature fusion, termed PCA-GASF-DCResNet. In this study, we first employ principal component analysis (PCA) for dimensionality reduction, followed by the Gramian angular summation field (GASF) transformation to convert the signal into 2D representations. Next, features are extracted via two parallel pathways using residual networks (ResNet) and subsequently fused for signal denoising. We evaluate the denoising performance of various algorithms and demonstrate that the proposed method exhibits superior performance under low signal-to-noise ratio (SNR) conditions. By applying the method to signals at multiple concentrations, we observe improved consistency and stability in both peak values and peak-to-valley ratios. Experimental results on measured <i>C</i><i>O</i>2 signals confirm that the proposed approach effectively suppresses noise interference, yielding smooth second-harmonic signals with reduced baseline fluctuations, thus facilitating accurate and reliable concentration inversion.