PSIV-A-6 - Computer Vision System to Predict Famacha© Degree in Sheep from Ocular Conjunctiva Images

Monday, June 27, 2022

1:30 PM – 2:45 PM CT

Location: Exhibit Hall A

Presenting Author(s)

LF

Luara Freitas, n/a

PhD student
University of Sao Paulo - University of Wisconsin-Madison
Madison, WI, United States

Author(s)

RF

Rafael Ferreira

Research Assistant
University of Wisconsin-Madison
Madison, WI, United States
RS

Rodrigo Savegnago

Michigan State University
Michigan, MI, United States
JD

Joao R. Dorea

Assistant professor
Department of Animal and Dairy sciences, University of Wisconsin-Madison
Madison, WI, United States
GJ

Guilherme J. J. M. Rosa, n/a

Professor
Department of Animal and Dairy sciences, University of Wisconsin-Madison
Madison, WI, United States
CP

Claudia Paz

Animal Science Institute
Sertaozinho, Sao Paulo, Brazil

Abstract: The Famacha© method has been created for detecting sheep unable to cope with the infection by Haemonchus contortus, through visual assessment of ocular conjunctiva coloration. The objective of this study was to develop a computer vision system (CVS) for prediction of Famacha© degree (FAM) from ocular conjunctiva images. The dataset consisted of 475 ocular conjunctiva images from animals in two farms. A Random Forest (RF) approach was used to segment the images, i.e. to select the pixels that belong to the ocular conjunctiva. After segmentation, the quantiles (0, 25, 50, 75, and 100%) of color intensity in each channel (red, blue, and green) were determined and used in multinomial logistic regression and RF models to classify the animals according to FAM. The classification methods were fitted using the information of quantiles, and the FAM (scores 1 to 5, non-anemic to severely anemic, respectively) as the target classes to be predicted. The RF presented the best performance in predicting degrees 1 to 4, with an F1 score, that represents the balance between the precision and the recall, of 50.0% for degree 1, 72.7% for degree 2, 75.4% for degree 3, and 62.2% for degree 4, where degrees 3 and 4 refer to animals that need treatment. None of the two prediction approaches were able to satisfactorily classify degree 5 animals. The results suggest that it is feasible to adequately predict the FAM in real-time from images of the conjunctiva ocular in sheep, providing key information for management decisions on the farm. Nonetheless, further studies and larger datasets may be needed to increase the classification accuracy of degree 5 animals. In conclusion, it was demonstrated that it is possible to successfully predict degrees of Famacha 1 to 4 in sheep via CVS using images collected in farm conditions.

Supported by the São Paulo Research Foundation (FAPESP) grants #2020/03575-8, #2018/01540-2 and #2016/14522-7, Brazil