The classification of Brain Tumor (BT) based on MRI images is one of the most crucial tasks in medical imaging because proper diagnosis directly influences the early treatment planning and patient survival. Although deep learning has made progress, it still experiences difficulties with accurate tumor segmentation, boundary preservation, and accurate differentiation of features across different tumor grades. In order to overcome these shortcomings, the proposed Dual Attention Holographic Convolutional Neural Network with Perfumer Optimization Algorithm (DAHCNNet-POA) offers a strong baseline of automated BT classification. The MRI images are first acquired on the BraTS 2018 dataset. Shape-Aware Mesh Normal Filtering (S-AMNF) is applied to perform preprocessing that removes noise without losing the structural and boundary integrity. A Visual Geometry Grounded Transformer (VGGT) is used to segment tumor regions accurately and capture both spatial and geometric relationships. After that, the discriminative features are obtained and classified in the DAHCNNet, which incorporates both spatial and channel attention to improve the learning of the representation. The network weights are optimized through POA, which enhances convergence and performance. The model got 99.12% accuracy of HGG, 99.18% accuracy of LGG, and Dice scores greater than 99, which proved its high reliability and effectiveness.