| dc.description.abstract | Skin diseases are among the most common health issues affecting pets, particularly cats, and are often difficult to identify visually without medical examination. This study aims to implement an image classification model based on transfer learning using the VGG16 architecture to detect skin diseases in pets. The model is designed to recognize six image classes: four types of skin diseases (dermatophytosis, scabies, pyoderma, and abscess), one healthy skin class, and an additional "unknown" class for irrelevant images. The dataset used consists of 10,000 images collected from public datasets and veterinary clinic social media, which were then categorized into six classes and divided into training, validation, and testing sets. The model was trained using limited fine- tuning and data augmentation techniques to improve generalization capability. Evaluation on the testing set resulted in an accuracy of 92%, with weighted average precision, recall, and F1-score of 0.92. The trained model was subsequently converted into TensorFlow Lite format and integrated into an Android application using Jetpack Compose. Performance evaluation was conducted on three Android devices with different specifications, showing an average inference time ranging from 1 to 6 seconds. Additional testing using 12 real-world images captured directly from mobile device cameras under varying lighting and clarity conditions achieved an accuracy of 75%. These results indicate that the proposed model is capable of accurate and efficient image classification and can be used as an early diagnostic support tool for detecting skin diseases in pets through mobile devices. | en_US |
