NO_x is a major pollutant in exhaust gas and primarily comprises NO and NO₂, with NO being the main component. To meet the stringent requirements for high-precision detection of NO concentration, this study proposes an NO concentration inversion method based on ultraviolet (UV) differential spectroscopy. This method employs an empirical wavelet transform (EWT) to decompose the UV spectrum of NO and applies an adaptive Savitzky-Golay (ASG) filter to process the decomposed signal. The filtered signal is reconstructed using the inverse EWT (IEWT). Finally, a long short-term memory (LSTM) neural network performs the inversion to determine the NO concentration. The spectral signal of NO is first processed via the EWT-ASG method and subsequently analyzed using the LSTM neural network. Results demonstrate that the predicted NO concentration has maximum and minimum errors of approximately 6% and 0.01%, respectively, when compared to the true values. Moreover, the root mean square error of the inversion accuracy improves by 35.86% compared to the unfiltered state. Thus, the proposed method provides strong technical support for the concentration inversion of NO and other components of exhaust gas.