注：深圳市科技计划项目（No.JSGG20191129105842769）；
    国家自然科学基金（No. 62003047）；
    北京市组织部骨干人才项目（No. 9111923202）
作者：祁文博，牟涛涛，陈少华
单位：北京信息科技大学仪器科学与光电工程学院，北京 100192
中图分类号：O433.4         
文献标识码：A        
文章编号：1006-883X(2022)03-0016-07
收稿日期：2022-02-10

摘要：使用便携式拉曼光谱仪检测矿物颜料具有检测速度快、操作简单、非破坏性等优势，但是会引入更多的环境光和噪声，为光谱分析带来困难。为了解决上述问题，提出一种深度卷积神经网络模型（Deep Convolutional Neural Network，Deep-CNN），该模型使用多个卷积层和多个池化层在深度方向提取拉曼光谱特征，然后使用全连接层进行分类，可以有效消除环境光和噪声的影响，自动、准确地识别三元混合矿物颜料的成分。实验结果表明：当训练集和验证集数量不少于140条光谱，或测试集信噪比不低于30 dB时，模型准确率为100%；使用与训练集成分不同的混合物光谱进行测试，准确率也达到了100%，这说明模型具备很强的鲁棒性。该研究为使用便携式拉曼光谱仪检测矿物颜料提供了一种可行的解决方案。
关键词：拉曼光谱；矿物颜料；卷积神经网络；便携式拉曼光谱仪

Composition Analysis of Color Mineral Pigments Based on Convolutional Neural Network and Portable Raman Spectrometer  
QI Wenbo, MU Taotao, CHEN Shaohua
(Beijing Information Science and Technology University, School of Instrument Science and Optoelectronic Engineering, Beijing 100192, China)
Abstract: Using portable raman spectrometer to detect mineral pigments has the advantages of fast detection speed, simple operation and non-destructive, but it will introduce more ambient light and noise, which brings difficulties to spectral analysis. In order to solve the above problems, the paper puts forward a kind of deep convolutional neural network (Deep-CNN) model. The model uses multiple convolutional layers and multiple pooling layers to extract raman spectral features in the depth direction, and then uses the fully connected layer to classify, which can effectively eliminate the influence of ambient light and noise, and automatically and accurately identify the components of 3-element mixed mineral pigments. The experimental results show that when the number of training and validation sets is not less than 140 spectra, or the SNR of test sets is not less than 30 dB, the model accuracy is 100%. Using the mixture spectrum different from the training set, the accuracy also reached 100%. This shows that the model has strong robustness. The study provides a feasible solution for the detection of mineral pigments by portable raman spectrometer.
Key words: raman spectral; mineral pigment; convolutional neural network; portable raman spectrometer

阅读全文

备注：2022年 第28卷 第03期