Research On Spectrum Processing And Analysis Algorithms Of Chromatoggraphy Workstation

Chromatogram has become the most commonly used separation and detection methods, widely used in various fields. Chromatograms have complex structure, multi-frequency components and overlapping peaks, making the analysis of chromatograms quite difficult. Thus, it is of great significance to research chromatogram processing including:noise filtering, baseline correction, peak detection and overlapping peaks resolution.The presence of noise affects the detection of peaks, thus affecting the calculation of peak area, reducing the accuracy of the chromatographic analysis, so noise filtering is the priority of chromatographic analysis. Generalized cross validation is used to obtain the optimal threshold, unlike classical threshold selection methods, which doesn't depend on the estimation of noise variance. Chromatographic signal is denoised by wavelet transform with layered optimal threshold, signal to noise ratio improved significantly.Due to the unstable operating conditions and other factors, baseline drift may be generated in actual chromatograms, having impact on net peak area calculation. A method based on iterative wavelet transform is proposed to correct baseline, gradually reducing peak contributions remaining in the approximation coefficients, to avoid the phenomenon of waveform distortion when extracting baseline by single wavelet transform. Energy of wavelet approximation coefficients is the criterion to determine the end of iteration, and the decomposition level of discrete wavelet transform.Signal peak in a specific location of chromatogram is corresponding to a specific substance (component), the peak area related to the material content, accurate detection of the peaks is the key to obtain valid information in chromatogram processing. Continuous wavelet transform which has strong anti-noise property is used to extract peak inflexions and peak positions, achieving peak detection. For the overlapping peaks in chromatogram, neural network is introduced to establish pattern recognition model, using the characteristic points detected by wavelet transform, to get accurate sub-peak area or the original peak characteristic points.