SpectralWeight: a spectral graph wavelet framework for weight prediction of pork cuts

Abstract: In this paper, we propose a novel approach for the quality assessment of pork carcasses using 3D shape analysis. First, we make a 3D model of a pork half-carcass using a 3D scanner and then we take advantage of spectral graph wavelet signature (SGWS) to build a local spectral descriptor. Next, we aggregate the extracted features using the bag-of-geometric-words paradigm to globally represent the half-carcass shape. We then employ partial least-squares regression to predict the weight of pork cuts for the quality assessment of carcasses. Our results demonstrate that SpectralWeight can predict the weight of different pork cuts and tissues with high accuracy. Although in this study we evaluate the performance of SGWS for the weight prediction of pork dissection, our framework is fairly general and enables new ways to estimate the quality and economical value of carcasses of different animals.