Dermatological diseases represent one of the leading causes of clinical consultations in dogs, affecting over 30% of sick animals. The application of artificial intelligence techniques to support automated diagnosis through image analysis emerges as a promising tool to facilitate and scale veterinary care, especially in regions with limited access to specialists. Although widely explored in human medicine, such approaches remain incipient in the field of veterinary medicine. Addressing this gap, the present study proposes an effective computational solution to assist veterinary diagnosis by developing an adapted deep learning model. The proposed optimized model achieved strong performance across all evaluation metrics, reaching 93.79% accuracy, 89.22% precision, 98.18% AUC, and 87.30% F1-score. Our approach was available in three public datasets of canine dermatological diseases, combined to enhance sample diversity and increase model robustness. This work contributes to advancing computational tools for supporting clinical decision-making in animal healthcare.