Horses Identification Through Deep Learning Algorithms

In many horse-related events, such as competitions, auctions, international travel, or doping control, the need to determine the identity of all horses involved is a must to ensure transperecy or simple misidentification of the horses. To circumvent these problems, the possibility of creating a deep neural network to classify live footage of the horses seems promising, as it could potentially provide another layer of validation and be relatively hard to falsify compared to the methods currently used. This paper produced a dataset composed of 17 different horses and around 10500 images, and developed a network model, creating it based on the VGG-16 model, and gradually changing its complexity and adjusting its architecture in each iteration until reaching the final accuracy ranging from 97.15% to 99.89%.