The oxygen concentration detection method based on near infrared absorption spectrum at 760.88nm was used to realize in-situ, non-contact detection of the residual oxygen concentration in the vial in the open environment on the lamp detector. The method based on Wavelength-modulated tunable diode laser absorption spectroscopy technology （TDLAS/WMS） uses the principal component extraction method （PCA） to extract the main characteristics of the second harmonic in the open light path, and then utilizes the genetic algorithm （GA） to optimize the BP neural network to build a concentration inversion model. This method can reduce the data required for calculation, suppress noise and improve the processing speed of post-processing data. The experimental results show that the average relative error of this method is reduced from 8.32% to 1.12%, and the coefficient of determination is increased by 8.86%, compared with the least square fitting method using semi-peak area. Compared with the average relative error of the single PCA-BP neural model, the mean relative error is reduced from 3.80% to 1.12%, and the coefficient of determination is increased by 2.81%.This method can effectively suppress the signal random disturbance caused by the open light path environment and improve the accuracy and stability of the detection of oxygen residual concentration in the vial.