Non-Destructive Monitoring of Moisture and Plastid Pigments in Tobacco Leaves During Curing Using near-Infrared Spectroscopy (NIRS) with Feature Fusion and a One-Dimensional Convolutional Neural Network (1D-CNN)

During the curing of tobacco leaves, dynamic variations in moisture and plastid pigment content are critical for regulating the progression of curing and optimizing operational procedures. To achieve accurate and nondestructive prediction of these parameters, a monitoring framework was developed in this study based on one-dimensional convolutional neural networks (1D-CNN) integrated with multi-source feature fusion. Near-infrared (NIR) spectra, leaf images, and temperature data were synchronously collected across the yellowing, color-fixing, and drying stages of curing. Spectral data underwent preprocessing via first derivative transformation, followed by feature wavelength selection using competitive adaptive reweighted sampling (CARS). Image color features were extracted from color-space descriptors, while texture attributes were obtained through the gray-level co-occurrence matrix (GLCM). Subsequently, NIR spectral features, image features, and temperature data were concatenated to capture complementary chemical, visual, and thermal information. Four distinct 1D-CNN quantitative prediction models, each based on different feature combinations, were developed and compared against conventional machine learning approaches. The results showed that models incorporating fused features outperformed those relying on single-feature inputs and traditional algorithms in terms of predictive accuracy and precision, achieving root mean square error (RMSE) values of 4.51%, 1.20 µg/cm2, and 0.0835 µg/cm2 on the test set, respectively. The fusion of NIR spectral, image, and temperature features significantly enhances the performance of predictive models for moisture and plastid pigment content during tobacco leaf curing, offering a valuable reference for the implementation of accurate, rapid, and nondestructive monitoring of chemical constituents in tobacco leaves.