<p>Diabetics’ major cause of blindness is retinopathy. Timely detection improves diagnosis and prediction. Hemorrhages accurately identify diabetes retinopathy (DR). Microaneurysms (MAs) and blood vessels affect retinal fundus hemorrhage. Therefore, it is essential to extract all these retinal features for correct medical interpretation. Deep Learning (DL) approaches enhance DR classification accuracy but fail to interpret retinal features. Handcrafted methods can analyze retinal characteristics for medical evaluation, but DR classification accuracy is lower. This study presents enhanced retinal feature extraction algorithms for DR classification. We proposed lightweight MAs and thin blood vessel extraction algorithms since fundus hemorrhages are the best indication of DR. The proposed model includes pre-processing, feature extraction, and classification. Fundus picture quality is improved via RGB image filtering with contrast augmentation in pre-processing. A combination of machine vision algorithms efficiently extracts tiny MAs and thin blood arteries. The extracted MAs and thin blood vessels are extracted and combined to build the hybrid feature vector for the DR classification. The extracted features are normalized to reduce the classification errors using the min-max normalization. Finally, different classifiers are trained using different datasets for the performance evaluation of DR classification. Four distinct classifiers, namely Artificial Neural Network (ANN), Support Vector Machine (SVM), AdaBoost, and Deep Neural Network (DNN), undergo training. This study empirically examines the suggested model utilizing two publicly accessible retinal datasets. The experimental findings demonstrate the superior efficiency of the proposed model in comparison to the underlying solutions. The proposed model demonstrates an 8.48% improvement in overall accuracy and a 42.38% decrease in classification time compared to the handcrafted methods.</p>