A prior knowledge boosted CNN-LSTM prediction model for oyster freshness evaluation based on an electronic nose

Abstract

Oysters are highly nutritious, but due to their high moisture and protein content, they are particularly prone to spoilage under unstable temperature conditions, posing significant challenges for food safety and supply chain management. Typical freshness detection methods widely used now show accurate result, but are destructive, time-consuming and costly, such as total viable count (TVC), total volatile basic - nitrogen (TVB-N), gas chromatography - mass spectrometry (GC-MS), and biogenic amines (BAs) testing. To address this issue, an oyster freshness detection method is proposed based on electronic nose (E-nose) technology combining with a novel freshness classification model named PriorBoost-CNN-LSTM. The relationship between E-nose sensor responses and typical physicochemical detection results was explored under varying temperature conditions and high related prior information of TVC, TVB-N, BAs with gas sensors were incorporated to the model to enhance the performance of E-nose system. This model is capable of predicting subtle freshness variations on an hourly basis under temperatures of 4 °C, 12 °C, 20 °C, and 28 °C, with accuracy rates increasing from 69.7 % to 81.2 % compared to the unboosted model. Experimental results verified a more efficient and precise solution for oyster shelf life management and the approach can be a good reference for other food’s freshness evaluation.