青砖茶压制压力优化及GCG近红外快速检测模型建立

青砖茶压制压力的选择至关重要，为探求压力与青砖茶品质及内含成分间的相互关系，并尝试对关键成分进行快速预测。以青砖茶为研究对象，设置了5个等级的压力值，通过感官审评和相关关系法分析了最佳压力值与品质和内含成分间的相关关系；应用标准变量变换、多元散射校正、一阶导数和二阶导数及组合方法进行降噪处理，应用反向区间偏最小二乘法筛选特征光谱区间并进行主成分分析，将主成分分别输入到3种信息传递函数的jump connection nets结构人工神经网络中建立定量分析模型。结果表明，最佳压力值为18 MPa；关键内含成分为：没食子儿茶素没食子酸酯（Gallocatechin Gallate，GCG）（P<0.05）；最佳预处理方法：多元散射校正+一阶导数组合方法；特征光谱区间：9 734.9～10 000，8 924.9～9 191.1，5 368.9～5 638.8，7 011.9～7 281.9，6 190.4～6 460.4，4 821.2～5 091.2，9 194.9～9 461.1，7 559.6～7 829.6，5 916.5～6 186.5 cm-1，前3个主成分累积贡献率为97.82%，以应用tanh传递函数建立的GCG人工神经网络模型结果最佳（Rp2=0.980，RMSEP=0.027），并有较好的实际应用效果（Rp2=0.948，RMSEP=0.041）。研究结果为其它重量规格青砖茶产品的研发和品质的快速检测奠定了理论基础。

Optimizing processing pressure of qingzhuan tea and development of GCG models for near infrared spectroscopy detection

Processing pressure is an important parameter during the production of Qingzhuan tea, and also for the identification of tea quality. Taking 100g Qingzhuan tea as the research object, this study aims to optimize the processing pressure, and then establish a quantitative model of key components, including pressure, quality and contents of Qingzhuan tea, using the near infrared spectroscopy. The processing pressure was set as 3, 6, 12, 18 and 24 MPa in the test. The sensory evaluation and correlation methods were used to analyze the relationship between the optimal pressure, the quality and the contents of Qingzhuan tea. Standard normal variate (SNV), multiple scatter correction (MSC), first derivative (FD) and second derivative (SD) and their combined methods were used to denoise the original raw spectrum during the preprocessing of data. Then, the backward partial least squares algorithm was used to select the characteristic spectral intervals, while the principal component analysis method was used to analyze them. Finally, the principal components were input into the jump connection nets structure artificial neural network (ANN) of three kinds of transfer functions, as linear [0,1] functions, logistic functions and tanh functions, respectively, to establish a quantitative analysis model. The results showed that 1) the optimum pressure was 18 MPa, while the content of gallocatechin gallate (GCG) was closely related to the pressure and the quality of Qingzhuan tea (P<0.05); 2) the optimum pretreatment method was MSC+FD method; 3) the characteristic spectral intervals were 9 734.9-10 000, 8 924.9-9191.1 cm-1, 5 368.9-5 638.8, 7 011.9-7 281.9, 6 190.4-6 460.4, 4 821.2-5 091.2, 9 194.9-9 461.1 cm-1, 7 559.6-7 829.6, 5 916.5-6 186.5 cm-1; 4) the cumulative contribution rate of the first three principal components was 97.82%; 5) the GCG artificial neural network model that established by tanh transfer function indicated the best results (Rp2=0.980, RMSEP = 0.027), with better practical application effect (Rp2=0.948, RMSEP=0.041). The findings can provide a theoretical foundation to develop more types of Qingzhuan tea products, and to rapidly detect their quality in tea industry.