关键词: 无基底扣除, 离散差分, 趋势累积, 遍历寻峰, 谱峰检测算法

Abstract:

The detection and analysis of spectral peaks play an important role in research on chromatography technology. However, in the process of collecting and transmitting chromatographic data, it is very difficult to detect spectral peaks owing to the interference of different levels of noise. Most of the traditional spectral peak detection algorithms follow three steps: spectral smoothing, baseline correction, and spectral peak recognition, which require high denoising and curve smoothing, and therefore increase the complexity of the algorithm. In addition, a traditional spectrum peak detection algorithm generally defines the shape of the spectrum peak by applying the base deduction method, and divides the spectrum peak into a single peak, overlapping peaks, and so on. Different detection methods are used for different types of spectral peaks, which lead to the shortcomings of traditional peak detection algorithms, such as high complexity, low automation, and susceptibility to distortion. Therefore, this study proposes a novel peak detection algorithm developed using a different point of view. The algorithm omits the base subtraction and spectral peak classification steps and instead detects spectral peaks directly based on the source data curve. In a traditional spectrum peak detection algorithm, the spectrum peak classification depends on determining a baseline. If the baseline is adjusted, the baseline will fit the spectrum peak more closely. At this time, the overlapping peaks can be regarded as two connected peaks. However, there is no so-called baseline in the source data curve, and therefore the proposed algorithm cannot classify the spectral peaks using the baseline approach. Instead, an obvious bulge or depression in the source curve is considered to be the spectral peak. This algorithm essentially performs three steps: discrete difference, trend accumulation, and searching for all peaks. First, the difference between adjacent data is obtained using a discrete difference process. The difference value is compared with 0, and either a 1 or -1 value is used to replace the difference value to reflect the data fluctuation trend. The signals representing the trend are accumulated, and the spectrum peak is located according to the sum of the accumulated signals. The algorithm uses three-point location; that is, the peak starting point, extreme point, and peak end point are used to describe the position of a spectral peak. Finally, according to the spectrum peaks obtained in the previous step, the magnitude of each peak is calculated, and the spectrum peaks are screened by a sorting method. In this manner, the algorithm skips the base subtraction part and obtains the spectrum peak directly. Therefore, to obtain the base part, the peak subtraction method is applied. This study used the C language to design and write the algorithm, and nitrogen adsorption and desorption chromatographic curves measured by several dynamic specific surface area analyzers were detected and analyzed. The results indicate that the proposed algorithm can accurately distinguish the peak part from the base part, and is robust to data curve burr, vibration, and other types of noise. The three-point location of the spectrum peak is very accurate and is not affected by its complex morphology. Therefore, it has strong universality. Compared with other algorithms, this algorithm has the advantages of accurate positioning, clear structure, and good stability and reliability. The application of the proposed peak detection methods such as base-free deduction and trend accumulation, in the adsorption and desorption chromatographic curve and has been proven effective in the determination of absorption and desorption chromatographic peaks.

Key words: no base deduction, discrete difference, trend accumulation, searching for peaks by traversing, peak detection algorithm