| Title |
Development of a Deep Learning-Based Model for Canine Skin Disease Diagnosis Using RandAugment and Patterned-GridMask |
| Authors |
김민준(Min-jun Kim) ; 박재범(Jae-beom Park) ; 조현종(Hyun-chong Cho) |
| DOI |
https://doi.org/10.5370/KIEE.2025.74.1.142 |
| Keywords |
Canine Dermatology; Deep Learning; Data Augmentation; Transformer Model; Medical Image Analysis |
| Abstract |
In contemporary society, pets are increasingly regarded as integral family members, contributing significantly to human quality of life. The growing prevalence of dog ownership has concurrently escalated the economic burden associated with veterinary care, particularly in managing common conditions like skin diseases. This study introduces an advanced deep learning-based diagnostic system for canine skin diseases, designed for practical application in home environments. We employed RandAugment to enhance data augmentation, thereby increasing the diversity of the training dataset. Furthermore, the implementation of Patterned-GridMask significantly improved the model's generalization capabilities. The use of the AdamW optimization algorithm was instrumental in mitigating overfitting, thus enhancing the model's overall learning efficiency. The proposed Transformer-based model, ViT/B-16, achieved an accuracy of 78.65% with the original dataset. With the integration of RandAugment and Patterned-GridMask techniques, the model's accuracy improved to 84.33%, underscoring its potential effectiveness for practical veterinary applications. |