Adsa Logo White Adsa Title White

Development of an identification system to recognize individual animals based on biometric facial features.

R. E. P. Ferreira

Events

06-23-2020

Join R. E. P. Ferreira on this page for a live text chat!
6:00 PM - 8:00 PM GMT

Abstract:

T85
Development of an identification system to recognize individual animals based on biometric facial features.
R. E. P. Ferreira*1, L. G. R. Pereira2,1, T. Bresolin1, G. J. M. Rosa1, J. R. R. Dorea1. 1University of Wisconsin-Madison Madison, WI, 2Embrapa Dairy Cattle Ju�z de Fora, MG, Brazil.

In livestock systems, animal identification and traceability are central for food security, data transparency, and consumer trust. A contemporary approach for animal identification is based on image analysis and computer vision, which in general exploit coat color differences among animals using color images. Such method, however, is not efficient for animal species and breeds with homogeneous color patterns. Point cloud deep learning has emerged as a promising machine learning technique that is able to extract features associated with 3D shapes. The objective of this study was to evaluate the efficiency of point cloud deep learning approach for animal identification using 3D images of dairy cattle faces. Images from 9 Holstein cows were acquired at the Dairy Research Cattle Center of the University of Wisconsin-Madison. The images were segmented to remove background and to extract animal faces. This approach was adopted to prevent potential bias toward an overoptimistic model due to recurrent environment noise present in the background of each animal image. A depth sensor (Intel RealSense model D435) was used to acquire 354 images, which were divided into train (n = 214), validation (n = 64), and test (n = 76) sets. A VoxNet point cloud deep learning network was implemented to generate the predictions. Data processing and analyses were implemented in MATLAB 2020a. The accuracy on the testing set was 75%, varying from 50% (for an animal with only 6 images used for training) up to 88.9% (for an animal with 33 images used for training). The results indicate that point cloud deep learning can be a powerful tool for animal identification based on their 3D biometric facial features. Such an approach can be adapted for application in other animal species with homogeneous coat colors and patterns.

Keywords: computer vision, pattern recognition, precision dairy farming.