Diabetic Retinopathy (DR) is a leading cause of vision loss among diabetic patients, and its early detection is essential to prevent irreversible blindness. In this work, we propose an automated framework for DR grade detection that combines deep-inceptional features with a Support Vector Machine (SVM) classifier. Retinal images are first preprocessed using median filtering, Gaussian filtering, and Contrast-Limited Adaptive Histogram Equalization to enhance contrast and highlight retinal lesions. Further, to address class imbalance and improve model generalizability, adaptive data augmentation is applied. After that, Deep-inceptional features are extracted from the augmented fundus images using a fine-tuned InceptionV3 network, which effectively captures hierarchical and discriminative information, including fine-grained lesion structures and retinal vessel patterns. The extracted features are then employed to train a SVM classifier using three distinct kernel functions: linear, polynomial, and radial basis function (RBF). Experimental evaluations on a benchmark dataset demonstrate that the SVM with the RBF kernel achieves better performance compared to the other kernel configurations. The proposed method outperforms several existing approaches in binary and multi-grade DR detection, showcasing its potential for effective and scalable DR screening.