High-throughput ratio and pressure analysis of mixed gas discharge emission spectroscopy based on deep convolutional neural networks

Optical emission spectroscopy is a powerful diagnostic tool for low-temperature plasma processes, yet the quantitative analysis of gas composition and pressure in mixtures remains challenging due to spectral line overlap and nonlinear effects. This study proposes a high-throughput analytical method based on a convolutional neural networks (CNN) to address this challenge. The method enables the simultaneous and accurate prediction of both gas composition ratios and total system pressure directly from a single emission spectrum of Ar/N2/O2 binary mixtures. A systematic dataset was constructed by collecting spectra under varied flow rates and pressures. An end-to-end CNN model was developed, and its performance was benchmarked against traditional machine learning methods, including random forest, support vector regression, and extreme gradient boosting. The CNN model demonstrated superior performance, achieving a test set root mean square error (RMSE) of <0.10 and R2 of >0.70 for gas proportion prediction, and an RMSE of ≈ 0.13 with R2 ≈ 0.70 for pressure regression. Interpretability analysis using Grad-CAM revealed that the high-importance wavelengths identified by the CNN align closely with the known characteristic spectral lines of the target gases. This alignment confirms the model's decision logic is physically interpretable and mitigates the “black-box” concern. The proposed CNN-based framework provides a reliable, efficient, and interpretable solution for intelligent online monitoring of plasma process parameters.