Breast cancer remains a major contributing factor to death from cancer in women worldwide, requiring accurate and prompt diagnosis. While many deep learning approaches have shown a lot of potential in automated histopathological image classification, existing methods often struggle with capturing fine-grained features and handling the high intra-class variability present in breast tissue images. This paper proposes TFEA-Net, a Triplet Feature Enhancement Attention Network that addresses these challenges through a novel attention mechanism combined with multi-scale feature extraction. The proposed architecture integrates a Swin Transformer backbone with a custom-designed model module that operates spatial and channel attention mechanisms in parallel. Multi-scale feature recognition is carried out through parallel dilated convolutions employing multiple dilation rates, allowing the network to identify discriminative features across different receptive field sizes. We evaluated TFEA-Net on the BreakHis dataset at different zoom levels (40X, 100X, 200X and 400X), achieving accuracies of 99%, 98%, 96% and 95%, respectively. The experimental results demonstrate that TFEA-Net outperforms the existing methods.