Quality classification of kiwifruit under different storage conditions based on deep learning and hyperspectral imaging technology

Kiwifruits contain various vitamins, amino acids, and minerals and are loved by consumers because of their sweet and soft taste. The storage time of kiwifruits after picking is short, and it needs to be stored in cold storage to prolong their maturation and softening time. In order to explore the effects of the storage environment on kiwifruit quality, hyperspectral imaging (HSI) technology was used to study the quality changes of kiwifruit under different storage conditions in the near-infrared (NIR) region. We proposed a deep learning-based extended morphology-nonlocal capsule network (EMP-NLCapsNet) algorithm to classify fruits stored at different temperatures [low temperature (4 °C, 75% relative humidity) and room temperature (18 ± 2 °C)] for different times (0, 2, 4, and 6 days). Extended morphological profile (EMP) and principal component analysis (PCA) are used in the EMP-NLCapsNet algorithm as the spatial and spectral feature extraction algorithms for kiwifruit hyperspectral images, respectively, and the extracted feature data blocks are fed into a non-local capsule network (NLCapsNet) to achieve classification. In addition, to further investigate the effect of storage time in the low-temperature environment on fruits, EMP-NLCapsNet was used to establish the association between hyperspectral deep features and fruits and classify the storage time for fruits. The classification map can visualize the difference between fresh and low-temperature stored fruits. The results showed that the combination of EMP-NLCapsNet and NIR-HSI could comprehensively, accurately, and rapidly differentiate the fruit quality of kiwifruit under different storage conditions, providing a new method for quality and safety testing in the fruit industry.