Deep learning-driven non-destructive quantification of chlorophyll and antioxidant activity in therapeutic plant...

Therapeutic plants are rich sources of bioactive phytochemicals and pigments that contribute to important biological functions, including antioxidant defense. Chlorophyll, the primary photosynthetic pigment, is closely associated with plant physiological status and contributes to the antioxidant potential of plant tissues. This study proposes a rapid and non-destructive method for the simultaneous quantification of chlorophyll a, chlorophyll b, total chlorophyll, and total antioxidant activity in powdered therapeutic plants using Fourier transform infrared (FT-IR) spectroscopy coupled with chemometric and deep-learning models. FT-IR spectra of ten medicinal plant species were collected and analyzed using partial least squares regression (PLSR) and a one-dimensional convolutional neural network (1D-CNN). For chlorophyll a, chlorophyll b, and total chlorophyll, the 1D-CNN models achieved coefficients of determination (R²) of 0.997, 0.998, and 0.997, with root mean square errors of prediction (RMSEP) of 0.13, 0.04, and 0.15 μg/mL, respectively. For antioxidant activity determined by DPPH, ABTS, and RPA assays, the 1D-CNN models yielded R² values of 0.994, 0.992, and 0.997 with RMSEP values of 1.52, 1.22, and 0.01%, respectively, accompanied by high ratios of performance to deviation (RPD), indicating strong predictive capability and robustness. In all cases, the deep learning models outperformed conventional PLSR models. These findings demonstrate that integrating FT-IR spectroscopy with deep learning provides a fast, accurate, and non-destructive approach for evaluating key quality attributes of therapeutic plant powders. The proposed method offers a promising alternative to labor-intensive and reagent-consuming conventional assays, enabling efficient quality control and high-throughput screening in the herbal medicine industry.