Deep learning proves its effectiveness in many applications, out of which computer vision is benefitted one of the most. Pre-trained models that have been trained and optimized for some other ap-plications in high end systems for several weeks can be reused and finetuned for some other context in the view to attain better outcomes than that provided by the deep learning networks built from scratch. Furthermore, combination of probabilities and predictions from these transfer learning models may yield still better outcomes than an individual transfer learning model. In this work, predictions from two transfer learning models namely Mobilenet-v2 and Inception-v3 are combined in the context of detection of Diabetic Retinopathy in retinal fundus images. In this regard, the proposal involves retinal image col-lection, retinal image pre-processing, feature extraction and classification through the combination of predictions from two transfer learning models, performance evaluation and evolving of best optimized weights for diabetic retinopathy detection. The methodology is evaluated on two popular datasets name-ly EyePACS and MESSIDOR-1. The methodology achieves comparable results when compared to that of the state-of-art techniques making it an apt choice for integrating it into real world applications help-ing the eye practitioners in early identification of this retinal condition.

convolutional neural networks, deep learning, diabetic retinopathy, fundus images, transfer learning