The analysis of moisture migration and distribution during nut drying process based on ODT-CNN by LF-NMR

Understanding moisture distribution in nuts is crucial for optimizing the drying process, a complex challenge since these internal changes are not externally visible. Low-field nuclear magnetic resonance (LF-NMR) can address this by providing both T2 relaxation signals to quantify moisture state and magnetic resonance (MR) images to visualize its distribution. However, methods that rely on a single data type are inherently limited: T2 signals provide quantitative detail but lack spatial context, while MR images provide spatial context but lack quantitative detail on water mobility. This trade-off leads to an incomplete and inaccurate assessment of the true dryness state. To overcome this issue, the One-dimension Transformer Convolutional Neural Network (ODT-CNN) is proposed, which a deep learning model designed to synergistically fuse these complementary data sources. The model features two specialized branches: a Transformer encoder captures long-range dependencies in the 1D T2 relaxation signals, while a CNN extracts hierarchical spatial features from the 2D MR images. The information from both branches is then integrated to achieve a holistic assessment. This fusion methodology achieved a state-of-the-art accuracy of 97.41%, significantly outperforming single-modality approaches. Furthermore, a water activity evaluation index based on the least squares method was proposed to assess the dryness of nuts also. This study highlights the potential of LF-NMR technology for analyzing water loss and distribution during nut drying, providing a reliable method and theoretical basis for accurately assessing nut dryness.