Detection and visualization of water content distribution in Chuju based on hyperspectral imaging technology and machine learning.

During the processing of dried Chuju (Chrysanthemum morifolium Ramat), rapid and non-destructive moisture content detection is crucial for quality control. Due to its complex structure, moisture distribution becomes highly non-uniform during drying process. However, few reports are available regarding the detection method, thus failing to meet practical production requirements. To address this challenge, this study employed hyperspectral imaging (HSI), combined with machine learning algorithms, to facilitate rapid moisture content detection and visualize its spatial distribution. Spectral data were collected from both the front and back sides of Chuju samples to develop a moisture content predictive model. Six preprocessing techniques and three machine learning algorithms were evaluated based on their performance. The results demonstrated that the combination of standard normal variate (SNV) with support vector regression (SVR) yielded optimal performance, achieving excellent prediction accuracy for both front and back side, with determination coefficient (R2p) reaching 0.986. Subsequently, pixelwise spectral analysis was performed to realize moisture distribution visualization. The findings indicated that during the drying process, moisture distribution was non-uniform; notably, the central region exhibited a significantly slower dehydration rate compared to the outer region. To ensure accurate prediction regardless of sample facing, four classification models were implemented to differentiate between the front and back sides. Among these models, linear discriminant analysis (LDA) achieved the highest classification accuracy at 98.74 %. Thus, this study provides a non-destructive and rapid method for effectively detecting moisture content in Chuju with high precision ensuring product quality, with the potential to be applied in other agricultural products.