Self-developed smart food dryer with integrated NIR sensor and spectral analysis for real-time moisture monitoring during drying

Efficient drying is essential for reducing post-harvest losses and ensuring food security, highlighting the need for smart drying systems that enable real-time monitoring and data-driven assessment of the drying process. In this study, a custom-designed smart food dryer integrating a near-infrared (NIR) sensor with conventional temperature, relative humidity, and air-velocity sensors was developed to facilitate real-time monitoring of the drying process and serve as a platform for future intelligent process control. The system was evaluated during the drying of apple slices by continuously acquiring NIR spectra in the 1555–1950 nm. Drying behavior was analyzed using thin-layer drying kinetic models, combined with advanced multivariate and spectral analysis techniques. The raw NIR spectra exhibited systematic variations corresponding to progressive moisture removal, while second-derivative spectra revealed a shift in the dominant water-related absorption band from approximately 1835 nm to 1870 nm, indicating changes in the water-binding environment during drying. Principal component analysis (PCA) effectively captured the drying trajectory, with PC1 scores showing rapid changes during the early drying stage (0–60 min) followed by stabilization after 90 min. PCA loading analysis identified key spectral contributions at 1665 and 1875 nm, suggesting distinct water environments and evolving water-matrix interactions. Two-dimensional correlation spectroscopy (2DCOS) further elucidated the sequential nature of spectral changes, with dominant responses in the 1880–1890 nm region confirming that water is the primary driver, while earlier variations in the 1630–1710 nm region indicated matrix-associated structural rearrangements preceding major moisture loss. Partial least squares regression (PLSR) models developed from raw NIR spectra accurately predicted moisture content in real time, achieving a prediction coefficient of determination (R2p) of 0.96 and a root-mean-square error of prediction (RMSEP) of 0.40 kg water/kg dry mass. PLSR loading and selectivity ratio analyses confirmed that the water-associated band around 1870–1880 nm was the dominant spectral feature governing moisture prediction during drying. Overall, the integrated smart drying system demonstrated robust, interpretable, and real-time monitoring of moisture during drying, highlighting its potential to support future intelligent process control, reduce post-harvest losses, improve process efficiency, and enable data-driven food preservation strategies relevant to food security.