Oxygen concentration is an important monitoring parameter in industrial process. Wavelength modulation spectroscopy of tunable diode laser absorption spectroscopy (TDLAS)was used to measure concentration of oxygen gas in industrial process by online monitoring. In this paper, we use the characteristic absorption peak of Oxygen at 760 nm to measure the oxygen concentration. Because of the strong coherence of laser, the detection sensitivity of TDLAS is severely restricted by optical interference noise. Especially at low concentrations, there is larger error by extraction signal in the absorption peak waveform because of the background fluctuation caused by optical interference. In response to this situation, Levenberg-Marquardt nonlinear fitting algorithm was proposed, and the use of the absorption line-derivative form of Lorenz line to fit the second harmonic signal and to extract the peak amplitude. On the other hand, Levenberg-Marquardt nonlinear fitting method needs a large amount of calculation. In order to develop the TDLAS analyzer can achieve real-time monitoring of the site, we use the C28 series of DSQ for data processing which support floating-point arithmetic, and the instrument achieve real-time monitoring capabilities in industrial process. Experimental results show that the algorithm can effectively extract the absorption peak characteristic value of the 2nd harmonic signal and overcome the background noise, The ratio of calculated by algorithm to actual oxygen concentration is nearly 1.01, the linear error of the concentration measurement is 1.18%.