Early spoilage detection of Hami melons using microbial diversity analysis and E-nose information dimension enhancement

Abstract

The problem of spoilage bacteria infection has become a topical issue in Hami melons, posing a potential food safety risk. Although traditional biochemical analysis methods have high precision and sensitivity, they are time-consuming and laborious. A novel method based on E-nose information dimension enhancement and convolutional neural networks was proposed for early spoilage detection of Hami melons. The succession patterns of the dominant spoilage bacteria in post-harvest Hami melons were systematically analyzed using high-throughput sequencing technology. Then, a Mantel-test model for the relationship between dominant spoilage bacteria and sensor signals was established to optimize the electronic nose array. Additionally, the one-dimensional time-frequency information of the E-nose was converted into two-dimensional images using the Gramian Angular Field (GAF) visualization. A ConvNeXt-AM network with channel-space feature enhancement was designed to evaluate the different degrees of moldiness of Hami melons, and the results were compared with those of classic ConvNet models (MobileNetV3, ResNet18, ShuffleNetV3, VGGNet, and ConvNeXt) and the Transformer model (MobileViT). The results showed that Aspergillus was the absolute dominant strain throughout the storage stage (relative abundance 45.93 %), and the Mantel-test model excluded the redundant sensors of W1C, W3C and W5C. In terms of signal characteristics, the time-domain signal was more sensitive than the frequency-domain signal. In the two-dimensional visualization features, the time-frequency domain images obtained from the GASF transformation were superior to those of the GADF. In integrating information, the ConvNeXt-AM model constructed from the time-frequency domain signals of GASF achieved the best performance, with an accuracy rate of 93.33 %. These findings provide a valuable reference for the early and non-destructive detection of post-harvest fungal diseases in fruits.