Improving Intramuscular Fat Assessment in Pork by Synergy Between Spectral and Spatial Features in Hyperspectral Image

Meat is a complex matrix of structural features exhibiting physical and chemical variations. The duality of the spatial and spectral information in the hyperspectral image of meat provides complementary information, and a synergistic fusion of the information will allow for the development of a rapid and non-invasive system based on hyperspectral imaging for assessment of a chemical component in meat. Intramuscular fat (IMF) is a critical factor in meat purchase decision making. Traditional techniques for IMF measurement are time-consuming destructive and laborious. This study investigated the use of data fusion techniques to fuse spectral and image data obtained from hyperspectral images of pork samples for the purpose of developing a technique for rapid and non-destructive prediction of IMF in pork. Following the acquisition of the images, image processing was conducted to create the region of interest. The mean spectral and the textural information data were obtained from the region of interest by spectra averaging and the use of Gabor filter and gray-level co-occurrence matrix techniques for image pattern recognition. These features were systematically fused using low-level, mid-level, and high-level data fusion techniques. The fused data were inputted into partial least square, and support vector machines to developed prediction models for IMF in pork. The result showed that the data fusion resulted in a higher prediction of IMF than the use of either spectral or textural information in isolation.