The 905 nm LiDAR is susceptible to Mie scattering interference in aerosol environments (e.g., smoke, fog, dust) during atmospheric detection, which severely degrades echo signal quality and restricts the detection accuracy and reliability of LiDAR in complex scenarios. Traditional denoising methods relying on fixed thresholds often suffer from over-smoothing of target signals or residual noise, as they fail to adapt to dynamically varying interference environments. To address these limitations, this study adopts an adaptive wavelet threshold denoising method based on high-low frequency energy ratio weights, aiming to balance efficient noise suppression and accurate target signal preservation.

An experimental platform was first constructed to measure target echo signals in natural aerosol environments, yielding valid noisy echo data at 148 m (short range) and 211 m (medium-to-long range). An innovative three-channel wavelet decomposition strategy was designed: the high-frequency channel was defined as the noise-dominated channel, the medium-frequency channel as the mixed channel of noise and signal features, and the low-frequency channel as the target core signal channel. To optimize threshold selection, an inter-layer consistency verification mechanism was introduced, utilizing the spatial correlation of wavelet coefficients across adjacent layers to distinguish signal and noise components. Additionally, an adaptive threshold adjustment strategy based on energy ratio weights was developed: interference intensity was quantified by calculating the energy ratio of high-frequency to medium-to-low frequency channels, and dynamic threshold factors were assigned for three typical scenarios (clean air, misty fog, thick smoke) to enable differentiated denoising. A six-layer decomposition was implemented using the sym4 wavelet basis, combined with the Garrote threshold function to avoid signal distortion caused by hard thresholds and over-smoothing induced by semi-soft thresholds.

Experimental validation and data analysis indicated that the proposed method outperformed traditional layer-wise threshold methods in both short-range and medium-to-long range detection scenarios. For 148 m short-range detection, the signal-to-noise ratio (SNR) increased significantly from 0.5 dB to 11.7 dB (an improvement of 9.6 dB, compared with traditional level method), the root mean square error (RMSE) decreased from 198.2 mV to 81.8 mV (a 51.6% reduction, compared with traditional level method), and the peak retention rate reached 94% (Fig.11). For 211 m medium-to-long range detection, the SNR rose from 2.6 dB to 18.7 dB (an improvement of 16.1 dB, compared with traditional level method), the RMSE dropped from 161.4 mV to 39.8 mV (a 69.7% reduction, compared with traditional level method), the peak retention rate reached 97%, and the target positioning error was controlled within ±1 m (Fig.14), demonstrating high detection accuracy. Further analysis revealed three core advantages of the method: the three-channel division adapted to energy distribution differences between noise and target signals across frequency bands, avoiding over-denoising or residual noise; inter-layer consistency verification leveraged the structural correlation of wavelet coefficients to accurately distinguish signal and noise components, enhancing threshold processing accuracy; the dynamic adjustment strategy enabled real-time adaptation to varying interference intensities, improving the algorithm’s robustness in complex aerosol environments. Compared with existing anti-interference technologies, this method exhibited superior performance in noise suppression and target signal integrity preservation, without requiring complex hardware modifications.

This study proposes and validates an adaptive wavelet denoising method based on high-low frequency energy ratio weights to address Mie scattering interference in 905 nm LiDAR atmospheric detection. By integrating three-channel wavelet decomposition, inter-layer consistency verification, and weight-based dynamic threshold adjustment, a balance between efficient noise suppression and accurate target signal protection is achieved. Experimental results confirm that the method significantly improves SNR, reduces RMSE, and maintains a high peak retention rate in both short-range and medium-to-long range detection scenarios. This work not only overcomes the limitations of traditional fixed-threshold methods but also provides practical and reliable technical support for anti-interference detection of LiDAR in complex atmospheric environments, with broad application prospects in fields such as autonomous driving and environmental monitoring.