Using deep learning and hyperspectral imaging to predict total viable count (TVC) in peeled Pacific white shrimp

In this study, deep learning method coupled with near-infrared (NIR) hyperspectral imaging (HSI) technique was used for nondestructively determining total viable count (TVC) of peeled Pacific white shrimp. Firstly, stacked auto-encoders (SAE) was conducted as a big data analytical method to extract 20 deep hyperspectral features from NIR hyperspectral image (900-1700 nm) of peeled shrimp stored at 4 degrees C, and the extracted features were used to predict TVC by fully-connected neural network (FNN). The SAE-FNN method obtained high prediction accuracy for determining TVC, with RP2 = 0.927. Additionally, TVC spatial distribution of peeled shrimp during storage could be visualized via applying the established SAE-FNN model. The results demonstrate that SAE-FNN combined with HSI technique has a potential for non-destructive prediction of TVC in peeled shrimp, which supply a novel method for the hygienic quality and safety inspections of shrimp product.