光谱学与光谱分析, 2019, 39 (12): 3755, 网络出版: 2020-01-07   

基于一维卷积神经网络的雌激素粉末拉曼光谱定性分类

Qualitative Analysis Method for Raman Spectroscopy of Estrogen Based on One-Dimensional Convolutional Neural Network
作者单位
1 燕山大学电气工程学院, 河北省测试计量技术及仪器重点实验室, 河北 秦皇岛 066004
2 燕山大学信息科学与工程学院, 河北省特种光纤与光纤传感重点实验室, 河北 秦皇岛 066004
摘要
拉曼光谱物质定性鉴别已被广泛应用于诸多行业和研究领域, 但传统拉曼光谱分析过程中的预处理主要依赖人为经验, 光谱特征提取虽然能够降低信号维度, 同时也会造成部分光谱信息损失。 特性相近物质本身光谱相似度较高, 受到测量过程中环境干扰和分析过程中多种误差影响, 导致最终分类效果并不理想。 针对此问题, 提出基于一维卷积神经网络(one-dimensional convolution neural network, 1D-CNN)的拉曼光谱定性分类方法。 实验采集雌酮(Estrone)、 雌二醇(Estradiol), 雌三醇(Estriol)三种不同雌性激素粉末的拉曼光谱, 设计随机平移、 添加噪声和随机加权三种光谱数据增强方法, 构建数量充足的拉曼光谱数据库用于神经网络模型训练与测试; 基于拉曼光谱数据特点提出一维卷积神经网络分类模型, 将光谱预处理、 特征提取和定性分类的全过程融为一体。 通过大量仿真实验, 优化所提出的神经网络模型超参数和训练过程并测试分类效果, 从预处理对光谱分类结果的影响和模型抗干扰性能两个方面与多种传统拉曼光谱分类算法对比, 评价模型性能。 实验结果表明, 本文提出的一维卷积神经网络模型可实现三类雌性激素粉末拉曼光谱快速准确分类, 分类正确率最高可达98.26%, 分析过程中无需光谱预处理和特征提取步骤, 简化了光谱分析流程, 并能保留更多有效信息。 同时, 当模拟测量噪声强度达到60 dBW时, 传统方法分类正确率均明显出现不同程度明显降低, 卷积神经网络模型依然能够取得96.81%的分类正确率, 说明相比对传统拉曼光谱分类方法, 所提出方法受光谱测量噪声影响更小, 鲁棒性更强, 适用于分析更复杂现场测量的强噪声拉曼光谱信号。 该研究结果表明深度学习方法在拉曼光谱的分析与处理领域具有很大的应用潜力和研究价值。
Abstract
The qualitative identification of Raman spectroscopy has been widely used in many industries and research fields, but the preprocessing in traditional Raman spectroscopy analysis mainly relies on human experience. Although spectral feature extraction can reduce the signal dimension, it also causes partial spectral information loss. Materials with similar characteristics have high spectral similarity. In addition, due to the interference of measurement environment and the various errors in the analysis process, the final classification accuracy is not ideal enough. Aiming at these problems, this paper proposes a novel Raman spectral qualitative classification method based on one-dimensional convolution neural network. The Raman spectra of three different estrogen powders, estrone, estradiol and estriol, were collected, and three Raman spectral data augmentation methods were designed to construct Raman spectral database; a one-dimensional convolution neural network classification model for Raman spectral data was proposed, which integrated the whole process of spectral preprocessing, feature extraction and qualitative classification. The hyper-parameters and training process of the proposed classification model were optimized and the accuracy was tested by simulation experiments. Experimental results indicated that the 1D-CNN model can classify three similar estrogen powders Raman spectroscopy with the highest classification accuracy of 98.26%. No spectral preprocessing and feature extraction steps were required in the analysis process, which simplifies the spectral signal analysis process and can retain more vital information. In addition, when the noise intensity of the simulated measurement reached 60 dBW, the classification accuracy of the traditional methods decreased obviously in varying degrees, but the 1D-CNN model could still achieve 96.81% accuracy. Compared with traditional Raman spectral classification methods, the proposed method was less affected by the noise of measurement process and had stronger robustness, which was suitable for Raman spectral signals with strong noise measured in more complex environments. The results of this study show that deep learning method has great application potential in the field of analysis of Raman spectroscopy.

赵勇, 荣康, 谈爱玲. 基于一维卷积神经网络的雌激素粉末拉曼光谱定性分类[J]. 光谱学与光谱分析, 2019, 39(12): 3755. ZHAO Yong, RONG Kang, TAN Ai-ling. Qualitative Analysis Method for Raman Spectroscopy of Estrogen Based on One-Dimensional Convolutional Neural Network[J]. Spectroscopy and Spectral Analysis, 2019, 39(12): 3755.

本文已被 1 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!