Abstract: The complex underground environment in coal mines, including factors such as light source fluctuation, noise, and environmental interference, can affect the Tunable Diode Laser Absorption Spectroscopy (TDLAS) system. This results in a decreased signal-to-noise ratio of the first-harmonic spectrum signal, which severely compromises the accuracy and stability of gas detection. To address these problems, a first-harmonic denoising algorithm for the TDLAS system based on an improved wavelet threshold was proposed. First, the optimal wavelet basis and decomposition level were determined using an enumeration algorithm to identify the optimal parameters suitable for the first-harmonic spectrum signal. Then, a continuously differentiable threshold function was developed to address the issues of abrupt changes in hard thresholds and detail loss in soft thresholds. Finally, an adaptive threshold based on the local variance of the first-harmonic spectrum signal was designed, enabling the threshold to dynamically adjust according to local signal characteristics, thereby achieving precise separation of noise and effective signals. Simulation results showed that compared with the traditional wavelet threshold algorithm, the improved wavelet threshold denoising algorithm increased the signal-to-noise ratio by 19.03 dB, reduced the mean square error by 98.75%, and improved the waveform similarity coefficient by 0.083 3, demonstrating superior denoising performance. Methane detection results indicated that the noise energy in the frequency band of the denoised signal was significantly reduced, and the useful signal was concentrated in the target frequency band, showing that the improved wavelet threshold denoising algorithm could effectively suppress the noise in the first-harmonic signal.