EXPRESS: Adulteration Detection for Poria Cocos Based on Near-Infrared Two-Dimensional Correlation Spectroscopy and Deep Learning.
L’essentiel
As a traditional Chinese medicinal material, the quality of Poria cocos (PC) significantly affects its pharmacological efficacy. However, adulteration of PC in the market has been frequently observed. Adulterated PC not only compromises its medicinal value but also poses potential risks to public health. Conventional identification methods are often labor-intensive, destructive, and insufficiently sensitive to complex adulteration scenarios. In this study, a rapid and nondestructive strategy for both qualitative and quantitative analysis of PC adulteration is developed by integrating near-infrared (NIR) two-dimensional correlation spectroscopy (2DCOS) with deep learning algorithms. NIR spectra in the range of 960-1600 nm are collected from pure PC and samples adulterated with 5 common adulterants at concentration gradients of 5%. Specifically, to enhance spectral resolution, a systematic 2DCOS dataset is established by combining 5 representative spectral preprocessing methods with 2 types of 2DCOS. Furthermore, a bidirectional long short-term memory network with an attention mechanism (BiLSTM-Attn) is proposed to fully exploit the sequential dependency of correlation features embedded in 2DCOS maps. Compared with experiments involving convolutional neural network (CNN), gated recurrent unit (GRU), long short-term memory (LSTM), and bidirectional LSTM (BiLSTM) models, the proposed BiLSTM-Attn achieved 100% classification accuracy and superior regression performance, with a mean coefficient of determination (R2) of 0.9920, a low root mean squared error (RMSE) of 2.41, and a high mean residual predictive deviation (RPD) of 12.85. Extended experiments demonstrate that the combination of NIR 2DCOS and deep learning algorithms enables accurate and robust detection of PC adulteration. The proposed framework offers valuable support for quality control in traditional Chinese medicine, and has a wide range of applications.
Synthèse détaillée
Résumé original
As a traditional Chinese medicinal material, the quality of Poria cocos (PC) significantly affects its pharmacological efficacy. However, adulteration of PC in the market has been frequently observed. Adulterated PC not only compromises its medicinal value but also poses potential risks to public health. Conventional identification methods are often labor-intensive, destructive, and insufficiently sensitive to complex adulteration scenarios. In this study, a rapid and nondestructive strategy for both qualitative and quantitative analysis of PC adulteration is developed by integrating near-infrared (NIR) two-dimensional correlation spectroscopy (2DCOS) with deep learning algorithms. NIR spectra in the range of 960-1600 nm are collected from pure PC and samples adulterated with 5 common adulterants at concentration gradients of 5%. Specifically, to enhance spectral resolution, a systematic 2DCOS dataset is established by combining 5 representative spectral preprocessing methods with 2 types of 2DCOS. Furthermore, a bidirectional long short-term memory network with an attention mechanism (BiLSTM-Attn) is proposed to fully exploit the sequential dependency of correlation features embedded in 2DCOS maps. Compared with experiments involving convolutional neural network (CNN), gated recurrent unit (GRU), long short-term memory (LSTM), and bidirectional LSTM (BiLSTM) models, the proposed BiLSTM-Attn achieved 100% classification accuracy and superior regression performance, with a mean coefficient of determination (R2) of 0.9920, a low root mean squared error (RMSE) of 2.41, and a high mean residual predictive deviation (RPD) of 12.85. Extended experiments demonstrate that the combination of NIR 2DCOS and deep learning algorithms enables accurate and robust detection of PC adulteration. The proposed framework offers valuable support for quality control in traditional Chinese medicine, and has a wide range of applications.