Recognition of Bovine Infectious Keratoconjunctivitis using Thermographic Imaging and Convolutional Neural Networks

Abstract

Infectious bovine keratoconjunctivitis (IBK) is considered the most important ocular disease in cattle rearing, causing significant losses in both the affected herd and as for producer. Because it is an infectious disease, the forms of diagnosis need to be improved. Currently, the diagnosis for IBK is performed through the evaluation of clinical signs by a specialist and confirmed by laboratory tests, which is usually a costly and time-consuming task. In this work, the use of infrared thermography for the acquisition of images of the bovine ocular region of healthy and infected animals by the IBK is evaluated, using this image base in the training and validation of a set of convolutional neural networks (CNN) with the aim of identifying whether or not the animal is infected in new samples. A total of 4.938 thermographic images of the bovine ocular region were used in the training and validation process of five different architectures of CNN, which were evaluated using cross-validation. The best results obtained in this study indicate that CNNs are able to correctly classify IBK clinical signs in thermographic images with an accuracy rate close to 80%.