Freshness Discrimination of Chilled Pork with Fluorescent Indicator Labels Based on Lightweight Convolutional Neural Network
CHEN Danni, ZHU Lei, WANG Lin, GAO Xiaoguang, ZHU Chenxin, DENG Wenjing, CHEN Bochao
1. College of Food Science and Biology, Hebei University of Science and Technology, Shijiazhuang 050018, China; 2. Hebei Shuangge Food Co. Ltd., Shijiazhuang 050021, China
Abstract：Color indications of fluorescent labels for freshness provide an important tool for monitoring meat quality in real time. This study developed a fluorescent label based on a zein film modified with rhodamine B (RhB) and fluorescein isothiocyanate (FITC). In the label, green fluorescence from FITC acted as a response signal, and red fluorescence from RhB as a reference signal. This fluorescent label exhibited dual emission responses when exposed to amines, FITC fluorescence increased whilst the fluorescence of RhB was undisturbed. The fluorescent label presented a clearly distinguishable color transition from pink to yellow-green, indicating significantly enhanced sensitivity and accuracy. Furthermore, convolutional neural network (CNN) was used to intelligently distinguish the color changes of the fluorescent label to reduce human visual errors. Lightweight CNN EfficientNetb0 was found to be superior to two other lightweight CNN (MobileNetv2 and ShuffleNetv2) and two non-lightweight CNN (ResNet50 and VGG16) in terms of discriminant effectiveness, with a recognition accuracy of 95.6%. The parameters and floating-point operations per second (FLOPs) of the EfficientNetb0 model were 4.01 MB and 0.398 GMACs, respectively, which achieved the best balance between FLOPs and accuracy. Therefore, this model can meet the need for the fast, accurate and nondestructive identification of chilled pork freshness. The research results provide a theoretical reference for the intelligent grading of the freshness of chilled pork using fluorescent indicator labels during cold storage and cold-chain transportation.
2024, Vol. 38 
