Pattern Filter Applied to the Noise Elimination of Chromatographic Signal
模式滤波器在色谱信号除噪中的应用

In this paper, a novel filter based on pattern recognition is proposed. The method consists of four major stages, namely:signal disintegration and labeling (SDL), pattern discrimination(PD), noise location(NL), and signal reconstruction (SR). In the SDL stage, chromatographic signals(CSs) are disintegrated into and labeled as a series of pattern primitives (PPs), namely:terms, flats, segments, etc.After a series of investigation, the noise patterns usually encountered in CS were summarized and classified into the following modes, i.e., spike, buzz, glitch, and ramp. Each of these modes is found to be formed by a particular combination and sequence of different pattern primitives. Based upon these concepts, special parsing procedures were developed to identify pattern primitives, and corresponding algorithms were developed to determine the starting and ending positions of each mode of noise pattern. By the end of NL stage, the starting and ending positions and signal intensities of identified noises are obtained.In the SR stage, an interpolation algorithm was used to delete or eliminate all of the identified noises and then reconstructed the CS. A series of sample CSs were used to test the behavior of the proposed method in comparison with other non linear digital filters. It was found that the proposed method gave little distortion upon chromatographic peak information and few unrecognized noises.