Abstract: Aiming at the problem that the ultraviolet-visible (UV-Vis) spectroscopy water quality detection system is susceptible to the noise interference from the instrument itself and the external environment, and there are a large number of system and stray light noise in the measured spectrum data, based on the analysis of the noise source of the water quality detection system by UV-Vis spectroscopy, a denoising method using genetic algorithm to optimize wavelet threshold is proposed and compared with the wavelet soft threshold, SG smoothing and median filtering methods. In order to evaluate the denoising effect, denoising experiments are carried out on the UV-Vis spectrum data of potassium hydrogen phthalate standard solution with the same concentration. The genetic algorithm is used to select the wavelet optimal threshold for denoising, and at the same time, the traditional wavelet soft threshold denoising, SG smoothing denoising and median filter denoising are also used for comparison. In order to verify the practical feasibility of the algorithm, the four methods are further used to denoise the spectra of actual water sample from a domestic sewage and outlets of a sewage treatment plant. The experimental results show that the wavelet threshold denoising effect based on genetic algorithm is obvious. Compared with the traditional wavelet soft threshold denoising, SG smoothing and median filtering methods, the signal-to-noise ratio of the new denoising method has been improved by 2.2994, 5.7066 and 2.6155 dB respectively, the root mean square error has been increased by 0.0028, 0.0087 and 0.0033, and the peak signal-to-noise ratio has increased by 2.0837, 5.2569 and 2.7375 dB. It is shown that the wavelet threshold denoising based on genetic algorithm not only suppresses the noise in spectral data, but also improves the system accuracy, which provides a new solution for water quality spectral denoising by ultraviolet-visible spectroscopy.

Key words: ultraviolet-visible spectroscopy, water quality detection, genetic algorithm, wavelet soft threshold denoising