A novel direct absorption spectroscopy gas sensing system based on end-to-end deep neural networks was proposed for measurements of gas concentration. One-dimensional convolutional neural network and deep multi-layer perceptron were explored to measure the concentrations of methane and acetylene. The accurate measurement results for both gases demonstrated that deep neural networks based direct absorption spectroscopy technique can be reliably applied to different gas molecules. The developed gas sensing system achieved more precise concentration retrieval compared with that of wavelength modulation spectroscopy, and fast computation speed as well as robustness to noisy conditions, laser aging and circuit parameter variation simultaneously. The combination of deep neural networks and direct absorption spectroscopy provides new ideas for the further research of gas absorption spectroscopy.

提出了一种基于端到端深度神经网络的新型直接吸收光谱气体传感系统，用于测量气体浓度。探索了一维卷积神经网络和深层多层感知器来测量甲烷和乙炔的浓度。两种气体的准确测量结果表明，基于深度神经网络的直接吸收光谱技术可以可靠地应用于不同的气体分子。与波长调制光谱相比，开发的气体传感系统实现了更精确的浓度反演，计算速度快，同时对噪声条件、激光老化和电路参数变化具有鲁棒性。