Neural networks-integrated metal oxide-based artificial olfactory system for meat spoilage identification

A custom-built metal oxide-based olfactory sensing system was used to analyze the headspace from beef strip loins (M.Longissimus lumborum) stored at 4 °C and 10 °C. Area-based features were extracted from the raw signals using various signal processing techniques. Classification models using radial basis function neural networks were developed using the extracted features and performance tested using leave-1-out cross validation method. The developed models classified the beef samples into two groups; “unspoiled” (<6.0 log10 cfu/g) and “spoiled” (⩾6.0 log10 cfu/g) based on the microbial population. Maximum total classification accuracies above 90% were obtained for the samples stored at the two temperatures. Scaling the signals did have a positive influence in improving the classification accuracies obtained. Back propagation neural network prediction model using the pooled data (containing the area scaled feature) resulted in a R-squared of >0.70 between predicted and actual spoilage population from the 10 °C and 4 °C stored samples.