Classification of Packaged Vegetable Soybeans Based on Freshness by Metabolomics Combined with Convolutional Neural Networks

Abstract

Background/Objectives: Effectiveness of modified atmosphere (MA) packaging for the preservation of the freshness of vegetable soybeans was confirmed by using metabolomics combined with convolutional neural networks (CNNs). Methods: Stored under a low O2, high CO2 environment, the vegetable soybeans’ freshness was tracked through changes in hue angle on the surface of the crops and metabolite levels compared to those stored under normoxia. Results: MA packaging slowed respiration and reduced pectin decomposition, succinic acid oxidation, and fatty acid consumption, all linked to freshness maintenance. Using 62 key metabolite concentrations as inputs, CNNs classified vegetable soybean freshness into seven categories with 92.9% accuracy, outperforming traditional linear discriminant analysis by 14.3%. Conclusions: These findings demonstrate MA packaging’s effectiveness in extending freshness of vegetable soybeans by monitoring specific metabolic changes. This will contribute to the advancement of research aimed at elucidating the relationship between freshness and metabolism in horticultural crops.