Abstract: The utilization of deep learning and transfer learning methodologies in the field of image classification has led to substantial advancements across a range of applications. This research paper presents an evaluation of the VGG16 convolutional neural network, enhanced through transfer learning and regularization techniques, for the specific task of sheep breed classification. The model was fine-tuned on a dataset of six sheep breeds, using preprocessing techniques like resizing, normalization, and augmentation. Architectural enhancements such as batch normalization, dropout, and dense layers helped reduce overfitting and improve generalization. A dropout rate of 0.5 with a batch size of 16 achieved the highest test accuracy of 80.00%. Higher dropout rates (e.g., 0.8) resulted in underfitting and lower performance. Overall, the use of transfer learning and dropout regularization significantly improved classification accuracy.
Keywords: VGG16, Transfer learning, Regularization Techniques, CNN, Sheep Classification.
|
DOI:
10.17148/IJARCCE.2025.14683
