A Rapid and Nondestructive Quality Detection Approach for Yongchuan Xiuya Tea Based on NIRS and siPLS-ANN Method

The efficient and non-destructive evaluation of Yongchuan Xiuya tea quality represents a key advancement in the tea industry. Near-infrared spectroscopy (NIRS), a non-invasive analytical technology, allows for the acquisition of spectral data while preserving sample integrity. Through preprocessing the spectral data and employing the synergy interval partial least squares (siPLS) method to identify characteristic spectral regions, principal component analysis (PCA) is applied, followed by the development of a Jordan–Elman artificial neural network prediction model (ANN) for tea quality assessment. The optimal spectral preprocessing approach identified in this study is a combination of multiplicative scatter correction and second derivative processing. Key spectral intervals include 4377.6 cm−1–4751.7 cm−1, 4755.6 cm−1–5129.7 cm−1, 6262.7 cm−1–6633.9 cm−1, and 7386 cm−1–7756.3 cm−1, with the first three principal components achieving a cumulative contribution rate of 99.05%. Utilizing a tanh activation function, the model exhibited strong predictive performance: an Rp2 of 0.980 and RMSEP of 0.341 for prediction set samples, and an Rp2 of 0.978 with RMSEP of 0.373 for unknown samples. These findings demonstrate the potential of integrating NIRS with Jordan–Elman neural networks for rapid and accurate Yongchuan Xiuya tea quality evaluation, establishing a solid technological foundation for the application of NIRS in tea quality assessment.