Banca de DEFESA: JONAS CARNEIRO ARAÚJO

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : JONAS CARNEIRO ARAÚJO
DATE: 01/09/2022
TIME: 14:00
LOCAL: SALA VIRTUAL - PLATAFORMA GOOGLE MEET
TITLE:

Video Image analysis on prediction of carcass characters of Santa Inês lambs

KEY WORDS:

Machine learning. Shape descriptors. Side view.

PAGES: 89
BIG AREA: Ciências Agrárias
AREA: Zootecnia
SUBÁREA: Produção Animal
SPECIALTY: Manejo de Animais
SUMMARY:

The objective of this work was study and develop evaluation methods through morphometric measurements and automatically with video image analysis (VIA) of slaughter weight, hot and cold carcass weights, commercial cut weights and tissue components, ribeye area and subcutaneous fat thickness of live lambs. Information and images of 92 Santa Inês and crossbred lambs from commercial herds, were used, as well as 67 information on carcasses Santa Inês and crossbred lambs, from commercial herds. Morphometric measurements were performed on cold carcasses in order to predict commercial characteristics of the carcass, meat cuts and meat quality. The most important measures were selected by the generalized regularized canonical correlation statistic (RGCCA), after this selection, the variables with the greatest weight were used to generate equations, through the multivariate adaptive regression splines (MARS) regression algorithm, for the prediction of carcass, primal cuts and meat characteristics. In addition, images of the dorsal and lateral view of the live animal were obtained and processed by the ImageJ 1.05i program, to obtain biometric measurements which were compared with measurements obtained directly from the live animal to see if measurements made by VIA can be as effective for prediction as measurements performed in vivo and if the use of live weight improves the prediction results, the elastic net algorithm was used to obtain the prediction equations. Then the images obtained in the side view were processed again in an automated way by the google collab platform where the descriptors were obtained in a way delimiting regions of interest in the live animal, the elastic net and MARS learning algorithms were used to find out which had the best fit for the prediction of slaughter weight characteristics and carcass information through the use of shape descriptors. Thus, through the work performed, we observed that the use of morphometric measurements performed on the cold carcass allows the prediction of carcass characteristics, commercial cuts and texture, the use of VIA in the image of the live animal to obtain biometric measurements are so effective how much measurements obtained in vivo allowing the prediction of characteristics such as body weight and hot and cold carcass weight as well as the weight of commercial cuts, AOL and EGS and the use of the google collab platform allows to obtain descriptors automatically thus, it is possible to predict information on animal slaughter weight and carcass weight and commercial cuts through an image obtained in the side view of the live animal, thus being a precision and automation tool for the rural environment.

BANKING MEMBERS:
Presidente - 035.432.296-64 - ANDRE GUIMARAES MACIEL E SILVA - UFPA
Interno - 1487106 - CRISTIAN FATURI
Externo à Instituição - FERNANDO HENRIQUE MELO ANDRADE RODRIGUES DE ALBUQUERQUE
Externo à Instituição - LUCIANO FERNANDES SOUSA - UFT
Externo à Instituição - LUIGI FRANCIS LIMA CAVALCANTI
Externo à Instituição - MARLON MARTINS DOS REIS