Transfer-Ensemble Learning Based Deep Convolutional Neural Networks for Diabetic Retinopathy Classification

In this research endeavor, our primary area of focus revolves around the classification of diabetic retinopathy (DR) into five different categories. We have devised an ensemble approach that harnesses the collective capabilities of two widely recognized pre-trained convolutional neural networks, namely Inception V3 and VGG16. Our innovative method is geared towards capitalizing on the unique strengths of these individual networks in order to elevate the performance of DR classification. To construct our ensemble model, we have judiciously chosen to freeze specific segments within the layers of both Inception V3 and VGG16. This strategic decision is aimed at fully leveraging the learned features within these networks. Our ensemble technique has been meticulously trained on the APTOS dataset., which is a repository of diabetic retinopathy images. We have meticulously divided this dataset into distinct training and validation subsets. In the training phase, our model is honed to adeptly classify retinal images into their respective DR classes. The results are evident in the experimental evaluations derived from the test data, which unequivocally affirm the efficacy of our ensemble technique for diabetic retinopathy classification. Notably, our approach achieves a remarkable accuracy rate of 96.40%.