Implementation Of Transfer Learning In Cat Breed Detection Using Web-Based Convolutional Neural Network (CNN)

Abstract

Cats are one of the most popular pets because of their friendly and adorable nature. Along with the increasing cat population in Indonesia and the variety of breeds, cat breed identification is a challenge in itself, especially for cat lovers to the animal conservation community. This research aims to develop an image-based cat breed classification system using transfer learning method with MobileNetV2 Convolutional Neural Network (CNN) model. This model was chosen because of its ability to produce high accuracy with good computational efficiency, making it suitable for use on devices with limited resources. The dataset used consists of 13,000 training images and 3,250 testing images of 13 cat breeds. The model training process was carried out up to 50 epochs with the addition of fine-tuning for 10 epochs, after previously terminating the process at the 60th epoch, resulting in a validation accuracy of 98.67%. Model performance testing also showed high average results of evaluation metrics, namely precision of 91.38%, recall of 91.39%, and F1-score of 91.33%. Based on these results, it can be concluded that the application of MobileNetV2 transfer learning is able to classify cat breeds accurately and efficiently. The website made makes it easy for users to recognize cat breeds by simply uploading images, making it very useful for the general public, professionals, and cat enthusiasts.