目的 采用红外光谱技术结合机器学习算法建立牛膝Achyranthes bidentata炮制品类别与炮制程度的定性判别模型。方法 采集不同炮制品与不同炮制程度牛膝的中红外光谱（mid infrared spectroscopy，MIRS），运用BP神经网络（back propagation neural network，BPNN）、遗传算法优化BP神经网络（GA-BP）、随机森林（random forest，RF）、径向基神经网络（radial basis function network，RBFN）、卷积神经网络（convolutional neural networks，CNN）等机器学习算法建立牛膝炮制品类别与炮制程度的定性判别模型；采集不同炮制品与不同炮制程度牛膝的近红外光谱（near infrared spectroscopy，NIRS），使用TQ Analyst软件中的判别分析法建立牛膝炮制品类别与炮制程度的定性分析模型。结果 机器学习算法模型结果显示CNN判别模型较优秀，BPNN、RF及RBFN性能相近，GA-BP模型性能相对较差。3个NIRS定性模型结果显示验证集准确率均为100%，可准确预测炮制品类别与炮制程度。结论 通过红外光谱技术建立的定性分析模型可作为牛膝炮制品类别与炮制程度的鉴别手段。同时提供了快速、无损的检测手段及可靠的数据分析方法，为中药材炮制品类别与炮制程度精准识别提供新的方法参考。

Objective To establish a qualitative discrimination model for the type and degree of processing of Niuxi (Achyranthes bidentata, AB) using infrared spectroscopy and machine learning algorithms. Methods The infrared spectra of AB with different processing types and degree was collected, and various machine learning algorithms, including back propagation neural network (BPNN), genetic algorithm-optimized BP neural network (GA-BP), random forest (RF), radial basis function network (RBFN), and convolutional neural networks (CNN) were used to establish a qualitative discrimination model for the type and degree of processed products of AB. The near-infrared spectra (NIRS) of AB with different processing types and degree was collected, and TQ Analyst software was used to establish a qualitative analysis model for the type and degree of processed products of AB. Results The results of the machine learning algorithm models showed that the CNN discriminative model was superior, the BPNN, RF and RBFN had similar performance, and the GA-BP model had relatively poor performance. The three NIRS qualitative models had validation accuracies of 100%, indicating that they could accurately predict the type and degree of processed products of AB. Conclusion The qualitative analysis model developed in this study by infrared spectroscopy can be used as a means to identify the type and degree of processed products of AB. It also provides a rapid and non-destructive means of testing and a reliable method for data analysis, with view to providing a new method of reference for the accurate identification of the type and degree of preparation of Chinese herbal processed products.

R283.6

国家重点研发计划“中医药现代化研究”重点专项项目（2018YFC1707000）；河南省中医药科学研究专项课题（2022ZY1156）