Intelligent prediction of black tea withering degree using grad-CAM-CNN assisted VNIR-colorimetric sensor array

The accurate prediction of the withering degree is essential for ensuring the aroma quality of black tea. In this study, we present an innovative intelligent platform that integrates a carefully designed colorimetric sensor array (CSA) with visible near-infrared (VNIR) spectroscopy, powered by an interpretable deep learning model (convolutional neural network, CNN) to predict withering degrees. Specifically, eight dyes were selected from sixteen candidates based on correlation analysis (R² > 0.36). These dyes were drop-cast onto filter paper to create the CSA, which was used to capture key volatile organic compounds (VOCs) released during the withering of tea leaves. Spectral data from the CSA were collected and modeled to identify the different withering stages. To further enhance model interpretability, gradient-weighted class activation mapping (Grad-CAM) was applied to highlight the spectral regions most relevant to classification. When compared to conventional models, the Grad-CAM-CNN demonstrated superior performance, achieving 99% accuracy and a computational efficiency of 0.0559 s/100 samples. This approach not only ensures the aroma quality of black tea but also provides an efficient, scalable, and cost-effective solution for real-time process control in tea manufacturing, with potential for adaptation to other food systems.