激光诱导击穿光谱技术与卷积神经网络相结合的中药材产地识别研究

Origin identification of Chinese medicinal materials combining laser-induced breakdown spectroscopy and convolutional neural network

为了更好地对道地药材产地进行识别,文中提出一种激光诱导击穿光谱(Laser-induced breakdown spectroscopy,LIBS)技术与卷积神经网络(Convolutional neural networks,CNN)相结合,并在网络结构中加入卷积块注意力模块(Convolutional block attention module,CBAM)的药材产地识别混合模型(CNN-CBAM).该模型采用端到端的网络结构,利用CNN挖掘数据中的深层特征,通过加入注意力机制来提升网络的特征提取能力.实验采集5个不同产地黄芪的LIBS光谱数据,通过构建的混合模型对测试集的识别精度进行评估,发现相较于未改进的CNN模型以及传统机器学习中的支持向量机和随机森林算法模型,改进后的CNN在测试集上的准确率可达到100%.研究结果证明了LIBS技术结合CNN-CBAM网络模型对中药材产地进行准确识别的有效性.

In order to better identify the origin of authentic medicinal materials,a model is proposed based on the laser induced breakdown spectroscopy(LIBS)combining with convolution neural network(CNN)and adding convolution block attention module(CBAM)into the network structure.The model adopts end-to-end network structure by using CNN to mine deep features in data,and adding attention mechanism to improve the feature extraction ability of the network.The LIBS spectral data of Astragalus membranaceus from five different habitats were obtained in the experiments,and the recognition accuracy of the test set was evaluated by using that hybrid model.Compared with the unimproved CNN model and the traditional machine learning support vector machine and random forest algorithm model,it was found that the accuracy of the improved CNN on the test set can reach 100%.The results showed that LIBS technology combined with CNN-CBAM network model is effective in accurately identifying the origin of Chinese herbal medicines.