Monitoring the red meat freshness by using combined dielectric spectroscopy and image processing

Regarding the importance of quality of meat and other daily consuming food stuffs in the growth and health of human society, development of quality diagnosing and monitoring systems for food materials are being paid increasing attention by investigators. In this study, 40 beef samples were subjected to macroscopic imaging and dielectric power spectroscopy at 20 frequencies in the range of 5-100 MHz during five days of storage at 5 ° C. It was hypothesized that combination of the two sensing methods would result in more information on physicochemical changes of meat during ageing. For any beef sample, 42 attributes (i.e. 20 dielectric variables including dielectric power at different frequencies and 22 texture and color features of the image) were extracted. Classification analyses for the day of storage were performed with five algorithms of neural networks including multi-layer perceptron (MLP), multinomial logistic regression (MRL), functional trees (FT), logistic model trees (LMT) and Bagging aggregation. The results showed that the dielectric power at different frequencies decreased with the storage day from e.g. 250 µW at 5 MHz on the first day to 100 µW at the same frequency on the fifth day. The results showed that image parameters of beef were more effective in classification than dielectric variables but combining the information of the both sensory techniques, after reduction using PCA, resulted in classification accuracies of %78 for functional tree (FT) algorithm and %77 for Bagging classification with MLP as the base classifier