Automated Dog Breed Recognition Using Cnn-Based Transfer Learning

Abstract

Image classification remains a major challenge in supervised machine learning, requiring the accurate assignment of images to specific categories. This study tackles this issue by leveraging transfer learning through the InceptionResNetV2 model, pretrained for multi-class dog breed classification. Transfer learning, a cornerstone in deep learning, enables the reuse of knowledge gained from one task to enhance performance on a related one. Instead of building a neural network from scratch, this approach adapts learned features—particularly the pretrained model weights—for the specific task of dog breed identification and also for Environmental safety.

The research employed the Stanford Dog Dataset, which includes 20,580 images spanning 120 dog breeds. Following preprocessing and data augmentation, several pretrained convolutional neural network (CNN) models—including VGG16, ResNet50, and InceptionV3—were fine-tuned for the task. Performance evaluation was conducted using key metrics such as accuracy, precision, recall, and F1-score. Among the models tested, InceptionV3 delivered the best performance, achieving an accuracy of 91.2%, thereby highlighting the efficacy of transfer learning in addressing fine-grained image classification problems like dog breed recognition.