Hybrid Method to Identify Diabetic Retinopathy

Convolutional Neural Network (CNN) is a deep learning method that performs well in the image data processing. The disadvantage of CNN is that it takes a long time for training and requires a lot of computer memory, so in this study, it is proposed to use the Hybrid method (Convolutional feature learning and Extreme Learning Machine classification) to overcome these problems. The Hybrid method Convolution Extreme Learning Machine (CELM) will classify fundus images of Diabetic Retinopathy (DR). World Health Organization (WHO) recognizes that DR is a significant eye disease that causes blindness and requires special attention because this disease is increasing quickly. The processes carried out in this research are preprocessing (Cropping, Resize, and Augmentation) and classification using CELM. The feature learning process extracts features of the image using various CNN architecture and classified by KELM. The overall accuracy result is obtained by the CELM method, which reaches 99.95% of accuracy and the best architecture obtained on ResNet50 using 800 hidden nodes and it produces a short training time of 1,539 seconds.