拉曼光谱神经网络分类与样本扩充方法

Neural Network Classification and Sample Expansion in Raman Spectrum

拉曼光谱广泛应用于岩石矿物分析,而神经网络需要大量样本才能对拉曼光谱做到较高的分类准确率。本文提出了一种拉曼光谱的样本扩充方法,使用Gold反卷积算法、波峰波谷寻峰算法和函数拟合方法获取光谱中所有的拉曼峰信息,然后在其中加入随机值重构拉曼光谱用于扩充数据集。实验结果显示,样本扩充方法将卷积神经网络模型的分类准确率由95.7%提升至98.6%,因此在样本数据量极少的情况下依然可以使用样本扩充方法对岩石矿物样本准确分类且稳定性很高。

Raman spectroscopy is widely used in rock and mineral analysis,but the Neural Network needs a large number of samples to achieve a high classification accuracy of Raman spectroscopy.So,a sample expansion method of Raman spectrum was proposed in this experiment.Gold deconvolution algorithm,peak and trough search algorithm and function fitting method were used to obtain all Raman peak information in the spectrum,and then random values were added to reconstruct the Raman spectrum to expand the data set.The experimental results show that the sample expansion method improves the classification accuracy of the Convolutional Neural Network model from 95.7%to 98.6%.Therefore,the sample expansion method can still be used to accurately classify rock and mineral samples with high stability under the condition of very few sample data.