A near‐infrared CO2 detection system for greenhouse gas based on PCA‐DNN

Abstract Carbon dioxide (CO 2 ) gas is one of the main greenhouse gases. The detection of CO 2 gas content is of great significance to the study of the greenhouse effect. The CO 2 detection system based on the principle of tunable diode laser absorption spectroscopy (TDLAS) was demonstrated. A distributed feedback (DFB) laser with a central wavelength of 1580 nm was used as the laser source, multipass gas cell (MPGC) was used as gas cell, and indium gallium arsenic (IGA) photodetector was used to complete photoelectric conversion, and the original extracted second harmonic signal was smooth and denoised by wavelet transform. The signal‐to‐noise ratio (SNR) of the wavelet‐filtered spectrum improved from 6.88 to 13.87 dB, an improvement of 2.02 times. Principal component analysis was used to reduce the complexity of the data by compressing the spectrum from 800 dimensions to 2 dimensions. For the concentration inversion, the back‐propagation deep neural network (BP‐DNN) model was used to perform standard gas concentration step experiments and compared with the back‐propagation neural network (BPNN). The experimental results show that the BP‐DNN inversion of CO 2 concentration has improved computational accuracy and the root mean square error (RMSE) is 3.55 times lower than that of the traditional BPNN, showing favorable application prospects.