Relationship between attributes extracted from carcass image: Prediction of carcass tissue characteristics and meat quality of hair sheep lambs

Abstract

The objective of this work was to propose a methodology for extracting characteristics from shape and texture descriptors of hair lambs carcasses, obtained by digital image processing. Information from 88 cold carcasses of castrated male lambs was divided into three different databases by the views of the images considered and their respective delimitations, with the purpose of evaluating the explanatory quality of these types of captures, as follows: DB1 – “dorsal view database”, aggregating the dorsal view delimitations (dorsal whole carcass + carcass transverse segmentation in dorsal view) with 64 Manifest Variables (MVs) in the IMAGE Latent Variable (LV); DB2 – “sideview database”, with the side view delimitations (sidewhole carcass + carcass transverse segmentation in side view) and 64 MV's in the IMAGE LV; and DB3 – “complete database”, accounting for all dorsal and side delimitations and totaling 128 MV's in IMAGE LV. Regarding the validation of the structural model, referring to the part of the model that evaluates the relationships between the LVs, only the DB1 III and DB3 III models met all the necessary requirements. The PLS-PM was effective in explaining the causal relationships between the variables studied. All models studied obtained sufficient performance to validate the measurement model, with models DB1 and DB2 obtaining better predictive and explanatory performance, with the set of characteristics obtained from the image of the lateral carcass, plus the weight of the cold carcass and the carcass scores, latent variables PRIMAL CUTS and QUALITY.