Searching lost dog by breeds sorting using computer vision and deep neural networks

Mendoza-Montoya J.; Cuarite F.A.Y.; Quispe G.W.P.; Paz A.A. · Proceedings of the 2023 IEEE 3rd International Conference on Advanced Learning Technologies on Education and Research, ICALTER 2023 (2023) · DOI: 10.1109/icalter61411.2023.10372918

Resumen

The loss of a pet has a significant emotional impact on the owner and the pet. This sad experience generates worrying in the owner, while the pet gets stressed with fear and loneliness. Supporting owners to find their pets in a short time is crucial because of loss or probable death. Regarding social media advertisements as fallible methods, this research proposes a lost dog identification based on breed sorting classification using computer vision recognition. A convolutional neural network of 19 layers deep (VGG-19) was used to achieve this purpose. A dataset of 1706 images of five popular dog breeds in Perú was chosen (chihuahua, Yorkshire terrier, pug, Pastor Aleman, and Beagle), of which 668 were used to test and 971 to validate the model. The trained model achieved high accuracy, averaging 92% after 30 training epochs. In conclusion, using computer vision algorithms, this model-based convolutional neural network looks promising for finding lost dogs quickly based on their breed's recognition.

Instituciones: Universidad Tecnológica del Perú