Progressive and precise detection of peanut leaf diseases is vital for enhancing crop output, reducing pesticide overuse, and promoting sustainable agriculture. Recent Deep Learning (DL) approaches such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViT) have demonstrated promising results; however, their performance is often constrained by limited labeled datasets, class imbalance, and insufficient feature robustness. To tackle these issues, this study introduces an innovative hybrid architecture that combines a Deep Belief Network (DBN) with a ViT for the automated classification of peanut leaf diseases. The DBN is employed as an unsupervised feature learning module to extract robust and discriminative representations from raw leaf images, effectively capturing underlying disease patterns even with limited labeled data. These learned features are subsequently fused with the ViT architecture, enabling enhanced global context modeling through multi-head self-attention. Extensive experimentation on a curated peanut leaf disease dataset demonstrates that the proposed DBN-ViT model achieves a classification accuracy of 98%, outperforming conventional CNN-based and standalone transformer models.