Application of Fine-Tuning on Convolutional Neural Networks to Improve Classification Accuracy of Feline Skin Diseases

Abstract

Skin conditions are among the most frequent reasons for veterinary visits, yet they remain notoriously difficult to distinguish by eye alone. For the average pet owner or general practitioner, overlapping visual symptoms between diseases like ringworm and scabies often lead to diagnostic uncertainty. This study addresses this challenge by developing an automated classification system based on the MobileNetV2 architecture. By employing a two-stage transfer learning strategy, where initial feature extraction is followed by targeted fine-tuning of layers from index 100 onwards, we adapted a general-purpose model to the specific nuances of veterinary dermatology. Our results indicate a significant performance leap: while standard training struggled with the complexities of skin textures, the fine-tuned model achieved a validation accuracy of 92%. These findings suggest that fine-tuning is not just a technical optimization, but a necessary step in making deep learning a viable, accessible tool for real-world veterinary diagnostics.