The increasing complexity and scale of deep learning models in computer vision have led to significant concerns regarding their environmental impact, particularly due to the substantial energy consumption required during training. This paper introduces Energy-Aware Adaptive Transfer Learning with Augmentation methodology, a novel approach designed to address these sustainability challenges. This method integrates real-time energy monitoring into the training process, dynamically adjusting data augmentation intensity and fine-tuning strategies based on energy usage and model performance feedback. By applying the proposed method to benchmark datasets such as CIFAR-10 and ImageNet, as well as a custom domain-specific dataset, we demonstrate that it achieves substantial energy savings without compromising accuracy or F1-score. Compared to traditional training methods and other state-of-the-art approaches, the method maintains high performance and significantly reduces the carbon footprint of model training. This research contributes to the growing field of sustainable AI by offering a practical solution that balances technological innovation demands with environmental responsibility, setting a foundation for future advancements in green AI practices.