C.A.R.E. Dog Disease Identification using Vision Transformers and XAI

Abstract

"According to the researches carried out, the bond between human and dogs has a history of over 15 000 years. Dogs are often known as humans’ best friend. However, this relationship is not always a positive one. As we all aware, there are different diseases that dogs carry that can eventually be transmitted to human. Due to these diseases, and the attention that needs to be provided when taking care of a dog, people are reluctant to adopt dogs. This has also led to a rapid increase of stray dogs all over the world in the year 2024. This can be a thread to both animal welfare and public safety. As a reply to this problem, the author has come up with a solution to develop a mobile application to identify diseases of dogs using an image captured with a mobile phone camera using vision transformer and CNN, choosing MobileNetV2 as the base model. To take the app to a higher level, the author has added XAI in this research. This way both dog owners and people who are willing to adopt stray dogs can detect diseases in dogs. After testing the developed models, the skin disease detection, eye disease detection MobileNetV2 models were able to achieve an accuracy of 0.91% and 0.55% respectively. On the other hand, the ViT model achieved an accuracy of 0.95% and 0.65% respectively."