Application of handheld near infrared spectrometer in quality control of traditional Chinese medicine: Rapid screening and quantitative analysis of Lonicerae Japonicae Flos adulteration

[Display omitted] •Portable NIR-chemometrics system for rapid adulteration screening of herbs.•99.58% cross-validation accuracy in detecting adulterants with 100% in test set.•Excellent predictive model for quantitative analysis of adulteration levels.•Enhanced TCM quality assurance from identification to quantification.•Handheld NIR spectroscopy validates as a portable tool for on-site TCM testing. Traditional Chinese medicine (TCM) prescription, with its intricate formulations and nuanced compositions, is a cornerstone of holistic health practices. However, the expansion of the TCM market has led to a surge in herb adulteration, which significantly undermines the quality and safety of these medicinal products. A case in point is Lonicerae Japonicae Flos (LJF), a widely utilized herb for treating colds, which has been adulterated by the cheaper Lonicerae Flos (LF), thereby affecting its therapeutic effectiveness. Therefore, a method utilizing handheld NIR spectroscopy combined with chemometrics has been developed to provide a portable, real-time solution for the rapid and accurate detection and quantification of adulterants in TCM. By collecting NIR spectra from LJF, LF and adulterated samples (AS), we’ve established a spectral database enabling deep insights into the correlation between spectral features and sample compositions. Resultantly, a classification model with a 99.58 % cross-validation accuracy, reaching 100 % for test set, effectively identified adulterants. And further spectral similarity analysis and classification identification of samples with different adulteration ratios were carried out. The cross-validation accuracy under the optimal model reached 98.38 %, and the test set accuracy was 99.20 %. In addition, the study extends to quantifying different levels of adulteration, employing 20 standard adulterated samples across a 0–100 % adulteration gradient. Via data preprocessing, feature extraction, and regression techniques, the full concentration prediction models were developed, later refined by segmenting samples based on high and low adulteration ratios. Under the SGFD_CARS_PLS (Savitzky-Golay smoothing with the first derivative_competitive adaptive reweighted sampling_partial least squares) model, exceptional performance was achieved, with a R2p of 0.983, RMSEp of 3.402, and RPDp of 7.757 for the homemade adulterated prediction set. In conclusion, the application of this technology not only improves the efficiency and accuracy of sc