Prediction of Pork Meat Total Viable Bacteria Count Using Hyperspectral Imaging System and Support Vector Machines

Abstract

If the total viable count (TVC) of bacteria in meat outnumbers certain number, it will become pathogenic bacteria. The paper is to explore the potential of hyperspectral imaging system based on support vector machines (SVM’s) for the detection of TVC of bacteria in pork meat. After the hyperspectral reflectance images were acquired and pre-processed, stepwise discrimination method was then performed to determine the optimal wavelengths which can characterize the gross change of TVC of pork meat. The five optimal wavelengths (480nm, 525nm, 650nm, 720nm and 765nm) covered a relatively large spectral range and accounted for about 94% of the total contribution to TVC prediction. In order to predict the TVC of pork meat, least square support vector machines (LS-SVM) was adopted as the modeling method, also to render the LS-SVM to exhibit best performance, 2 inferences within Bayesian evidence framework were employed to optimize its parameters. The prediction model based on the optimal five wavelengths was able to predict TVC with r = 0.87 and the result was considerably better than that of ANNs and MLR method. This research demonstrated the feasibility of using the hyperspectral imaging system coupled with the modeling method based on LS-SVM is a valid means for nondestructive determination of TVC of pork meat.