Nondestructive intelligent and portable detection of postharvest translucency and internal browning in pineapples using visible/near-infrared spectroscopy

Pineapple internal browning manifests as darkened translucent spots in the central tissue, with the number and area of these spots progressively increasing during storage. Translucency is characterized by excessive water accumulation in the flesh, leading to tissue softening which increases susceptibility to mechanical damage. This study innovatively utilizes the penetration characteristics of visible/near-infrared spectroscopy to achieve real-time detection and onset time prediction of postharvest internal disorders in pineapples by comparing different preprocessing methods and modeling strategies. Furthermore, we propose incorporating local spectral feature data as a key indicator for translucency detection, combined with feature-extracted data to enhance detection accuracy. To address systematic batch variations, we employ direct orthogonal signal correction to eliminate irrelevant spectral information, thereby improving model generalizability. Experimental results show that the maximum accuracy of the pineapple translucency detection model reached 95.2 % (training set) and 94.3 % (validation set), respectively. The dual-batch detection model for internal browning achieved an accuracy exceeding 90 % in both the training and validation sets. Meanwhile, the prediction model for the onset time of internal browning achieved a maximum accuracy of 93.7 % (training set) and 90.4 % (validation set). This work establishes a novel nondestructive detection method for postharvest pineapple disorders.