Determination and visualization of moisture content in Camellia oleifera seeds rapidly based on hyperspectral imaging combined with deep learning.

Moisture content (MC) is crucial for the storage, transportation, and processing of Camellia oleifera seeds. The purpose of this study was to investigate the feasibility for detecting MC in Camellia oleifera seeds using visible near-infrared hyperspectral imaging (VNIR-HSI) (374.98 ∼ 1038.79 nm) coupled with deep learning (DL) methods. Firstly, a method was proposed that utilized particle swarm optimization (PSO) to search for the optimal hyperparameters (batch size and learning rate) in the convolutional neural network regression (CNNR) model. The prediction performance of various models including partial least squares regression (PLSR), support vector machine regression (SVR), AlexNet, and CNNR was compared using both raw spectral data and preprocessed data. Then, four feature extraction algorithms (successive projection algorithm (SPA), competitive adaptive reweighted sampling (CARS), PSO, and the optimal DL framework) were used to extract spectral variables. The optimal hybrid prediction model PSO-CNN-SVR was determined, with coefficient of determination (R) of 0.918 in prediction set. In addition, the optimal simplified model was used to generate spatial distributions to visualize MC in Camellia oleifera seeds. The study results showed that the HSI technique combined with DL provides a reliable and efficient approach for achieving non-destructive detection and visualization of MC in Camellia oleifera seeds.