Banca de DEFESA: FRANCINE OTILIA VOGEL

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : FRANCINE OTILIA VOGEL
DATE: 07/08/2020
TIME: 09:00
LOCAL: Web Conferência via Google Meet - meet.google.com/yvu-opgc-jjy.
TITLE:

Application for estimating the weight of beef cattle.

KEY WORDS:

Computer vision; Mask R-CNN; Precision animal science; Python; TensorFlow.

PAGES: 45
BIG AREA: Ciências Agrárias
AREA: Medicina Veterinária
SUMMARY:

The present work developed an application to estimate the body weight of beef cattle. The project evaluated 91 animals from three feedlots, all animals were male of Taurine breeds and their crosses, with an average weight ranging between 255 and 506 kg. The animals were weighed on an electronic scale without previous fasting. The measurement of the thoracic perimeter was performed caudally, the scapula passing through the sternum and the spinal processes of the thoracic vertebrae, using the thoracic weighing tape. After weighing, with the Samsung cell phone camera, model A5, the animals were photographed at an average distance of 5 meters. With the aid of the ImageJ program, body area, body length taken laterally between the ventral end of the shoulder and the tip of the ischium and the height of the back measured at the backline to the curve of the ridge behind the front leg were measured. The application was developed in the Python 3.7 language, using framework Kivy. Keras frameworks with Tensorflow were used to create the bovine identification model and the Mask R-CNN for image segmentation. The data were submitted to the Shapiro-Wilk normality test, Pearson's correlation test and multiple regression analysis with the method stepwise for detecting the best predictive model through the Akaike Information Index. Statistical analyzes were performed using the R statistical program and the level of significance adopted was 5%. The predictive model selected by the statistical analysis was the one that included only the body area as a predictor variable. It presented a correlation of 0.83 (p <0.01). In this sense, it is suggested to use this predictive equation: Body Weight = 293.2 + 0.27 * Body Area (r2 = 0.68; p <0.01). The estimate of the body weight of beef cattle by body area measured from the photographic image proved to be viable.

BANKING MEMBERS:
Presidente - 1760760 - CARLOS EDUARDO NOGUEIRA MARTINS
Externo à Instituição - DIEGO PERES NETTO - UFSC
Externo ao Programa - 1901309 - FERNANDO JOSE BRAZ