[Research on Denoising Ultraviolet Spectrum Signal with An Improved Effective Singular Value Selection Method]

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Signal Processing

Background:

• Spectral signals are prone to various noise sources, including thermal, mechanical, and random noise.
• Online monitoring systems necessitate denoising methods that minimize human operational dependency and parameter selection challenges.

Purpose of the Study:

• To develop an automated spectrum denoising method based on singular value decomposition (SVD).
• To improve the selection of singular values for effective noise reduction in spectral data.
• To evaluate the denoising performance on UV spectrum signals across varying signal-to-noise ratios.

Main Methods:

• A novel singular value selection method is proposed, defining a fuzzy area (θ1–θ2) based on singular value differences and integrated information.
• Fuzzy C-means clustering is employed to determine membership and assign weight coefficients to singular values within the fuzzy area.
• The SVD-based denoising method is applied to UV spectrum signals with different signal-to-noise ratios.

Main Results:

• The proposed method demonstrates a significant denoising effect, effectively restoring the original spectral signal.
• Quantitative evaluation using signal-to-noise ratio, root mean square error, normalized correlation coefficient, and smoothness ratio confirms the method's efficacy.
• The data-driven approach minimizes the impact of parameter selection, enhancing usability in online monitoring.

Conclusions:

• The proposed SVD-based spectrum denoising method offers a robust and automated solution for noise reduction in spectral detection.
• The improved singular value selection and fuzzy clustering enhance the accuracy and reliability of spectral signal restoration.
• This method is well-suited for online monitoring applications requiring efficient and operator-independent denoising.