To improve the accuracy of multi-component gas analysis in infrared spectroscopy and simplify the workflow, an infrared spectroscopy gas detection method based on an improved convolutional neural network is proposed. This method can not only identify a variety of gas categories but also finely identify the concentration of gas. To verify the model identification effect proposed in this paper, eight kinds of gases such as CH4 and C2H6 were used as the sample gases for gas identification and concentration classification, and the corresponding hardware was used to complete the development of the system. The experimental results show that the accuracy of the model method for gas species identification can reach 90%, and the accuracy for concentration identification is the same. In addition, compared with the traditional CNN method, the recognition effect is significantly improved. With the improvement of the data set, the number of gas categories detected by this method and the measurement accuracy will be improved.

为了提高红外光谱多组分气体分析的准确性并简化工作流程，提出一种基于改进卷积神经网络的红外光谱气体检测方法。该方法不仅可以识别多种气体类别，还可以精细识别气体的浓度。为了验证本文提出的模型识别效果，以CH4、C2H6等8种气体作为样气进行气体识别和浓度分类，并采用相应的硬件完成了系统的开发。实验结果表明，模型方法气体种类识别准确率可达90%，浓度识别准确率相同。此外，与传统的CNN方法相比，识别效果显着提高。随着数据集的完善，该方法检测的气体类别数量和测量精度将会提高。