Retinal vascular occlusion (RVO) is a severe eye condition that can result in significant vision loss if not detected and treated early. Deep learning (DL)-based models have been employed to identify vascular abnormalities in retinal fundus images, but their performance is often hindered by insufficient training data. To address this limitation, we propose an ensemble model that automates the detection of RVO using multiple pre-trained models, leveraging transfer learning for improved diagnostic accuracy. Additionally, data augmentation and preprocessing techniques are incorporated to generate synthetic images and enhance image quality, thereby improving the training process. Extensive experimentation on real datasets demonstrates that the proposed ensemble model outperforms existing approaches, achieving a precision of 97%, sensitivity of 95%, and specificity of 93%. These findings highlight the model’s potential for real-time deployment in RVO diagnosis, as it effectively integrates multiple transfer learning models to enhance overall performance.