Study on VGG16 Transfer learning Model for Goat/Sheep Image Classification

The utilization of deep learning and transfer learning methodologies in the realm of image classification has resulted in notable progress across diverse domains. This research paper presents an analysis of the performance of VGG16 model when incorporating transfer learning techniques in conjunction with various regularization techniques for the purpose of image classification of goat and sheep images. By employing the technique of transfer learning, we have successfully fine-tuned the VGG16 models on a dataset comprising six distinct classes of goats and sheep. This process has enabled the models to adapt their pre-existing features, which were initially trained on a different task, to the specific requirements of our target task. Performance metrics such as accuracy, precision, and recall rate are employed in the evaluation of a model's capabilities. Although each strategy had its own distinct advantages, the strategies that consistently yielded balanced precision, recall, F1-scores, and accuracy were a dropout rate of 0.5 and L2 regularization. Whereas L2 Regularization outperformed the other models with a testing accuracy of 0.839.