Fine-Grained Image Classification (FGIC) refers to the task of separating visually similar categories with slight variations. This is a challenging task due to high intra-class variability and inter-class similarities. Traditional methods have failed to properly address these challenges; thus, sophisticated methods are required for the proper classification of FGICs. The current research proposes a new ensemble attention-based transfer learning combination framework using EfficientNet B0EfficientNet-B0, DenseNet 121DenseNet-121, and ResNet 152ResNet-152 models augmented by Convolutional Block Attention Module (CBAM) modifications. Leveraging the respective strengths of these frameworks, the combination framework seeks to optimize feature extraction operations while prioritizing discriminating region identification over hurdles of varying scales and occlusions. Our suggested framework in comparison to other state-of-the-art models like Vision Transformers (ViT) and API-Net achieved a stunning 94.25% accuracy on the CUB-200-2011 dataset, surpassing ViT [1] as well as API-Net [36]. In contrast to Vision Transformers, Our suggested ensemble model forms a balance between efficiency and accuracy, and for that reason, it is best suited for Fine-Grained Image Classification (FGIC) on real-world applications. Through the use of K-fold cross-validation and attention mechanisms, the architecture guarantees generalizability and strength and offers an effective and scalable solution for high-precision FGIC tasks.