一种离子迁移谱谱图重构及特征峰提取算法

Ion Mobility Spectrometry Spectrum Reconstruction and Characteristic Peaks Extraction Algorithm Research

离子迁移谱(IMS)是一种常压下快速、高灵敏度的痕量化学物质检测方法,广泛应用于化学战剂、爆炸物和毒品等检测领域。在离子迁移谱定性和定量分析中,采集到的原始谱图除了包含样品自身信息外,还包含了电噪声、背景干扰等噪声,特别是当分析物浓度低时,噪声会严重影响定性和定量分析的准确性。为提高离子迁移谱技术化学物质识别准确率,需要对离子迁移谱谱图信号进行重构。本文提出一种可同时实现离子迁移谱谱图重构和特征峰提取的新方法。通过建立优化目标函数,采用l1范数作为线性惩罚项,λ为正则化参数用来调节惩罚项在优化过程中的比例。为了求解优化目标函数,首先构造一个由Gaussian分布函数构成的超完备字典来表示离子迁移谱离子特征峰峰形,采用替代函数方法对优化目标函数进行迭代求解,当达到重构谱图与原始谱图均方根误差小于设定的阈值时停止迭代。为了验证提出的方法性能,分别采用仿真数据和甲基磷酸二甲酯(DMMP)样本数据进行验证,其中仿真数据由高斯分布函数字典原子及高斯白噪声组成。与此同时,我们对仿真数据和真实样本数据分别采用小波软阈值、小波硬阈值及S-G平滑滤波算法进行去噪重构。采用均方根误差(RMSE)和信噪比(SNR)作为评价指标,实验结果表明该方法成功实现离子迁移谱谱图重构和特征峰提取,预处理结果比其他三种方法有显著的性能提升,为开展离子迁移谱定性和定量分析研究提供了基础。

Ion mobility spectrometry(IMS) is a rapid, highly sensitive analytical method for the gaseous samples with a low detection limit. It is widely used to detect chemical warfare agents, illegal drugs and explosives. The original spectrum contains not only sample information, but also noise. Especially when the concentration of the analyte is low, the accuracy of qualitative and quantitative analysis based on IMS technology is seriously influenced. It is necessary to reconstruct the spectrum before qualitative and quantitative analysis. In our article, a new method simultaneously achieved the spectrum reconstruction, and characteristic peaks extraction was proposed. In the optimization function, we chose l1 norm as the linear penalty. The regularization parameter λ was used to adjust the scale of the penalty in the optimization. Solve the optimization function, a Gaussian dictionary was constructed to represent the shape of peak firstly, and the surrogate function algorithm was adopted to solve it. When the root mean squared error between the reconstructed and original spectrum achieved the set threshold, the algorithm was stopped. To evaluate the performance of our method proposed, the simulated data set and DMMP sample data set were used. The simulated data set was composed of Gaussian functions and Gaussian noise. Meanwhile, we compared our method with wavelet using a soft threshold, wavelet using hard threshold and S-G smoothing methods. Root mean squared error(RMSE) and signal to noise ratio(SNR) were used to compare the results of different methods. The experiments results show that our method has significant improvement than other methods. Based on the proposed method, qualitative and quantitative analysis can be carried out.