Determining the quality level of ready to-eat stuffed mussels with Arduino-based electronic nose

Abstract

In this study, the performance of a pre-designed and low-cost Arduino electronic nose for determining the quality of stuffed mussels was analyzed. In addition, 1000 images were taken on each storage day in order to determine the quality levels of stuffed mussel groups with open and closed shells by machine learning. Freshness limit values of stuffed mussels were determined as 200 for MQ3 and MQ135 sensors and 100 for MQ9 on the 3rd storage day when the total viable count (TVC) value exceeded 3 log CFU/g. In the study, faster neural networks with lower prediction times, such as SqueezeNet and GoogLeNet, were compared with ResNet-50, ResNet-101 and DenseNet-201 neural networks, which have larger prediction times but better accuracy. Study data showed that residual network (ResNet) 50 and Teachable Machine (TM) had high success in determining the quality levels of stuffed mussels.