Quality Analysis Of Chinese Medicinal Materials Based On Machine Learning And XRF Spectrometer

With the continuous strengthening of drug supervision and management and the improvement of quality awareness in the domestic Chinese medicine industry,the quality control of Chinese herbal medicines and decoction pieces is also facing new problems and challenges.According to the 2020 edition of the "Chinese Pharmacopoeia",the adulteration of Chinese medicinal materials and the excessive heavy metals will affect the safety and effectiveness of Chinese herbal medicines,which should be highly regarded.The complex pretreatment and chemical analysis of samples are often necessary steps for the detection of traditional Chinese medicine elements,which consume a lot of manpower and material resources for experiments.X-ray fluorescence spectrometry(XRF)experiments only require simple pretreatment of samples,with short consumption time,high reproducibility,and rapid detection of Chinese herbal medicine elements.Relying on the general project of the National Natural Science Foundation of China,this research carried out in-depth research on the detection of authenticity and excess of heavy metals in Chinese medicinal materials based on XRF spectroscopy.The main research contents are as follows:(1)Obtain the experimental spectrum by XRF and remove the background and matrix effects.Background subtraction was performed using an iterative discrete wavelet-based transform,and elemental matrix effects were corrected with Compton normalization.Based on the comparative analysis of lead(Pb)element in the honeysuckle spectrum,it is proved that the algorithm can effectively eliminate the influence of spectral background and matrix effect,and improve the determination coefficient of the Pb element calibration curve from 0.77 to 0.91.(2)Research on the authenticity discrimination algorithm of honeysuckle based on Fisher's discrimination.The algorithm uses t-distributed random neighborhood embedding for feature dimensionality reduction,and then establishes a classifier model based on Fisher discrimination,with an accuracy rate of 96%.Compared with other commonly used classification algorithms,the experimental results show that compared with the K-nearest neighbor algorithm,the accuracy rate is increased by 6%;compared with the fuzzy C clustering mean algorithm based on calculating the membership matrix,the classification accuracy of the proposed algorithm is not affected.The influence of sample type is universal.(3)Research on the over-standard classification of honeysuckle heavy metal elements based on transfer learning-support vector machine(Tradaboost-SVM).Firstly,the traditional machine learning method support vector machine(SVM)and Adaboost model were used to classify four heavy metal elements.Secondly,in view of the lack of honeysuckle sample data,soil data and Monte Carlo simulation spectrum big data were used to expand the honeysuckle sample.Finally,the transfer learning algorithm was studied,and a detection and classification model of the heavy metal element Pb exceeding the standard in honeysuckle from different data sources was constructed.The results show that the average classification accuracy of the Tradaboost-SVM transfer learning model in the training set and test set is 97% and 94%,respectively,which is higher than the accuracy of the SVM model and the traditional Adaboost classification model.In conclusion,this study established an elemental detection method for Chinese medicinal materials based on XRF spectroscopy.Using Fisher's discriminant to test the authenticity of Chinese medicinal materials is more accurate and universal;using the transfer learning algorithm,an efficient,feasible and stable model transfer method between different samples is proposed,which improves the accuracy of heavy metal over-standard classification.It provides technical support for the authenticity identification of a large range of Chinese herbal medicines and the on-site diagnosis of heavy metals exceeding the standard in the future.