激光诱导击穿光谱结合深度学习在乌龙茶品质判别中的应用OA

[Objective]Tea quality assessment and authenticity verification are crucial aspects in the tea industry that significantly impact both producers and consumers.The production process of oolong tea is complex,the production cycle is long,and the quality is affected by many factors.It is of great significance to develop a rapid identification method for oolong tea quality to ensure the quality and value of tea.Traditional detection methods,while established,suffer from notable limitations including high subjectivity,time-consuming procedures,and the consumption of a large number of samples.These challenges have prompted the search for more efficient and reliable analytical approaches.This study aims to develop a rapid,small-sample,and non-destructive tea classification method by combining laser-induced breakdown spectroscopy(LIBS)with deep learning algorithms,specifically targeting the differentiation of oolong tea of different qualities.[Methods]Optimal experimental parameters for the tea matrix were first determined by systematically investigating the influence of excitation power(70-190 mW),tableting pressure(6-18 MPa),and detection delay time(0-440 ns)on the LIBS signal at 589 nm.The spectral intensity,signal-to-background ratio(SBR),and signal-to-noise ratio(SNR)of this line were used as evaluation criteria.The optimal conditions were identified as an excitation power of 150 mW,a tableting pressure of 15 MPa,and a delay time of 80 ns.Under the optimized conditions,LIBS spectra of tea samples were acquired across five selected wavelength regions(378.5-406.6 nm,502.0-529.9 nm,574.6-602.6 nm,631.5-659.4 nm,and 752.5-779.6 nm),which correspond to the emission lines of key elements(e.g.,Ca,Na,K,and CN)relevant to tea quality.Comparative analysis was conducted between principal component analysis(PCA)and residual neural network(ResNet)to evaluate their classification performance.Additionally,the study investigated the impact of time resolution on model performance to optimize the analytical parameters.Both full spectrum and segmented spectral analyses were performed to determine the most effective approach for tea classification.[Results]In the discrimination of tea qualities,the ResNet model demonstrated exceptional performance,achieving 100％accuracy in validation sets.At the same time,the classification accuracy of adulterated tea of different quality and pure tea also reached 100％.While full spectrum analysis showed superior performance compared to segmented spectral analysis,the latter still achieved classification accuracies generally above 85％when the sample size exceeded five.In addition,the classification accuracy at a delay time of 80 ns was higher than that at 60 ns,indicating that optimizing time resolution can help the model to screen for effective information to some extent.[Conclusion]The combination of LIBS and the ResNet model provides a rapid and accurate method for distinguishing tea qualities,offering an efficient approach for tea quality control and authenticity verification.This method requires minimal sample preparation,uses small sample quantities,and delivers quick analysis results.The classification accuracy remains consistently high across the optimal time resolution range for full spectrum analysis,demonstrating the model's stability.The findings of this study make important contributions to both the practical application of LIBS technology in tea analysis and the broader field of agricultural product authentication.The demonstrated capability to achieve accurate classification without complex sample preparation represents a significant advancement in the field of tea quality control.The success of methodology suggests potential applications in other areas of food authentication and quality control,where rapid,accurate,and non-destructive analysis is increasingly important.Using oolong tea from the same manufacturer as a sample is of great significance for quality control,market value,brand reputation and sales process monitoring of small-scale tea manufacturers in the tea production process.This discovery not only promotes the application of LIBS technology in the identification of tea grades(especially rare tea),but also helps avoid market problems such as inferior-quality tea and the mixing of genuine and counterfeit products to a certain extent.