Purpose: The aim of this work was to develop a rapid, cost-effective procedure for the detection of frozen-then-thawed minced beef using multispectral imaging and FTIR spectroscopy either separately or in parallel.
Methods: In this study, freshly-ground beef was purchased and divided in ~75g-portions on seven separate occasions. Fifteen samples per case were placed in Petri dishes, multispectral images of the first five were immediately acquired and then, ~3g-portions were used for FTIR measurements. The remaining samples were frozen (-20°C) and stored for 7 and 32 days (5 samples/case). Samples were thawed for 4h at 4°C and subsequently subjected to similar data acquisition. Data analysis methodologies, i.e. Partial Least-Squares Discriminant Analysis (PLS-DA) and Support Vector Machines (SVM), were utilized for model building. While five meat batches were used for calibration and internal validation, models were further validated with independent data from the last two batches to avoid over-optimistic results. In total, 105 multispectral images and FTIR spectra were collected, along with microbiological measurements per batch as background information.
Results: Results showed a clear separation of fresh vs. frozen samples, as all samples were classified correctly during calibration and validation in the case of SVM. However, it was more difficult to separate frozen samples depending on storage time.
Significance: In conclusion, this study proves that sensor data could be used in a rapid quality control/assurance system for the detection of frozen-then-thawed minced beef. While some studies have explored the use of sensor data previously, this study not only uses two different sensors, but also utilizes a validation scheme that proves its effectiveness when applied to independent data.