Non-Destructive Determination of Moisture Content in Husk-On Fresh Corn Using Multichannel Visible–Near-Infrared Spectroscopy Combined with Deep Learning

Using spectroscopic technology for the accurate and non-destructive determination of moisture content (MC) in husk-on fresh corn ( Zea maize L. sinensis Kulesh) is crucial for optimizing harvesting periods, ensuring quality, and maintaining nutritional value. However, corn husks interfere with the propagation of incident photons within corn kernels, leading to acquired spectral signals that contain information unrelated to the kernels themselves, thereby decreasing the accuracy of moisture detection in the kernels. This study developed a multichannel visible and near-infrared (Vis-NIR) spectral acquisition system based on spatially resolved diffuse reflectance technology for MC detection in husk-on fresh corn. The developed system mitigates the interference of husks on the acquired spectral signals by collecting spectral information from multiple detection positions offset at specific distances from the incident light source. Meanwhile, three model building strategies based on deep learning frameworks, including feature-level fusion, data-level fusion, and decision-level fusion, were proposed and compared. Results showed that the decision-level fusion model with standard normal variate (SNV) preprocessing achieved the highest prediction accuracy, with a coefficient of determination (R 2 p ) of 0.897 and a root mean square error of prediction (RMSEP) of 4.13%. Furthermore, multichannel data relatively enhanced model performance, with the four-channel combination achieving the best performance. This study demonstrates the potential of deep learning and multichannel spectral data fusion in improving MC prediction accuracy, offering a practical solution for non-destructive moisture measurement in fresh corn.