Salmon freshness detection based on dual colorimetric label indicator combined with lightweight CNN models

Given the widespread consumption and high perishability of salmon, real-time and accurate freshness monitoring is crucial for ensuring food safety and reducing economic losses. In this study, two pH indicators, curcumin (CUR) and alizarin (AL), were used to make a dual colorimetric label indicator with polyvinyl alcohol (PVA) and sodium alginate (SA) as the film-forming substrates. This indicator effectively detects quality changes during salmon storage and exhibits vivid color responses. By integrating the dual colorimetric label indicator with deep learning, we trained and tested the lightweight convolutional neural network (CNN) model MobileNetV2, achieving a detection accuracy of 98.59 %. Additionally, we selected SqueezeNet, a lightweight CNN with an inference time of approximately 14 ms, for real-time and high-throughput detection. To deploy these models, we developed a smartphone application (APP) with a user-friendly interface and multiple detection functionalities to meet the needs of professional and non-professional users in various practical scenarios. The dual colorimetric label indicator yields enhanced information and greater accuracy over the single-label approach. Moreover, lightweight CNNs offer novel ideas for freshness detection. Their integration with intelligent packaging through a mobile app enables rapid and non-destructive detection. The research results may hold significant potential in food safety and quality control.