A deep learning-enhanced electrophoresis method for rapid freshness monitoring in cold-stored turbot (Scophthalmus maximus)

Ruiyuan Liu, Yong Sun, Shanyu Wang, Nan Liu, Ling Zhao, Qi Liu, Rong Cao,
A deep learning-enhanced electrophoresis method for rapid freshness monitoring in cold-stored turbot (Scophthalmus maximus),
Microchemical Journal,
Volume 218,
2025,
115775,
ISSN 0026-265X,
https://doi.org/10.1016/j.microc.2025.115775.
(https://www.sciencedirect.com/science/article/pii/S0026265X25031236)
Abstract: Accurate and rapid assessment of fish freshness is essential for ensuring product quality and optimizing cold chain logistics. However, conventional methods—such as sensory evaluation, total volatile basic nitrogen (TVB-N), and K-value analysis—are often subjective, time-consuming, and costly. In this study, a novel freshness classification approach integrating electrophoresis with deep learning (DL) was developed using cold-stored turbot as a case study. Freshness indicators including TVB-N, total viable counts (TVC), and K-values were monitored throughout refrigerated storage. Simultaneously, electrophoresis images of the corresponding samples were obtained via paper electrophoresis. Based on a dataset comprising 855 electrophoresis images, four deep learning models, namely convolutional neural network (CNN), ResNet18, ResNet50, and Vision Transformer (ViT), were established, and their performance was systematically evaluated and compared. The ViT model achieved the best performance with a classification accuracy of 94.15 % and a ROC-AUC of 0.9967. Subsequently, prediction results of the optimal model were interpreted and justified with attention rollout method, which visualizes attention heatmaps across the four freshness categories. The proposed framework offers a sensitive, efficient, and scalable solution for fish freshness evaluation and provides a theoretical basis for the development of intelligent food quality monitoring systems.
Keywords: Deep learning; Electrophoresis; Freshness; Turbot; K-value; Vision transformer