Early and accurate identification of Maize Leaf Diseases (MLD) is critical for crop health and achieving maximum production. Automated diagnosis systems are becoming more valuable across today's agricultural sector. These systems assist farmers with early tolerance response treatment of MLD while protecting maize yield and quality. Existing methods struggle to reliably identify MLD for early infections appear as tiny dots and asymmetrical shapes in natural scene images. In this paper, a Multi-Scale Vision Transformer (MS-ViT) is proposed to quickly recognise MLD. Using Maize leaf dataset the spatial proximity and pixel intensity technique removes noise and preserve edges. Then smoothen the maize leaf image using Joint Bilateral Filtering (JBF) to develop high quality images. Subsequently, the processed image is given as an input to the Grabcut segmentation that segments the leaf area by graph-cut algorithm and Gaussian Mixture Model (GMM). After that MLD image is featured by Gray Level Co-occurrence Matrix (GLCM) find the energy, contrast, homogeneity, and correlation values for further analysis. An MS-ViT framework is proposed to enhance the classification of MLD image using maize leaf dataset. Using the proposed framework implemented python software, the approach achieves a higher accuracy of 95% compared to other techniques.