Data Fusion of Electronic Nose and Multispectral Imaging for Meat Spoilage Detection Using Machine Learning Techniques

Meat quality plays a significant role in the consumers' health condition; hence, the constant pursuit for techniques capable of objective and accurate quality assessment by the meat industry. Multispectral imaging and electronic noses are valuable techniques for the rapid and non-destructive detection of meat spoilage. In order to take advantage of the complementary information provided by these two different sensing devices, a high-level data fusion strategy was explored. Through this fusion scheme, the aim of this work is to estimate initially the population of total viable counts of Pseudomonas spp., Brochothrix thermosphacta and lactic acid bacteria, and then to categorize the status of the meat samples into three classes (fresh, semi-fresh, and spoiled). The issue of small size available datasets was addressed by generating additional "virtual" sample sets, through the use of neural networks. Neuro-fuzzy based regression models were implemented and their outputs were combined in order to estimate these microbiological populations. Following the evaluation of these estimations, it can be argued that the most efficient prediction was obtained through the fusion of these sensing devices, the coefficients of determination, the residual prediction deviation, and the range error ratio exceeded the 0.98%, 5.4%, and 14.73%, respectively. In parallel, the classification rate for the grouping of the testing samples into three classes was perfect. Based on the acquired results, the proposed analytical concept could potentially provide an alternative approach towards the efficient detection of meat spoilage.