The gas concentration is of great significance in the analysis of various fields. Because in the process of gas concentration detection , the impact of spectral dimensions leads to large errors in the detection results. In order to reduce the adverse effects in the measurement process , a gas concentration detection research based on near-infrared spectroscopy technology is proposed. . The near-infrared spectral baseline was corrected by castration - standard nor- mal transformation. Combine the generalized S transform and singular value to decompose and denoise the near-infra- red spectrum to improve the spectral quality. Based on principal component analysis ( PCA) , the partial least squares ( PLS) dimensionality reduction method is proposed for the dimensionality reduction of near - infrared spectroscopy. Based on Lambert Beer's law , Lorenz linear fitting of near-infrared spectral absorption line is introduced , and gradient descent method is used to directly fit the pre-processed near-infrared spectral absorption signal , and the final gas con- centration detection result is calculated. The experimental results show that the detection results of the proposed meth- od are basically consistent with the actual gas concentrations when detecting the concentrations of toluene , propane and propylene , effectively reducing the residual sum of squares and root mean square error , and the detection time is less than 2. 3 s.