THE APPLICATION OF HAAR WAVELET AND BACKPROPAGATION FOR DIABETIC RETINOPATHY CLASSIFICATION BASED ON EYE RETINA IMAGE

Abstract

Diabetic Retinopathy is a disease that attacks eyes retina and can cause blindness. The severity of Diabetic Retinopathy consists of four; they are; normal, Diabetic Retinopathy Non-proliferative, Diabetic Retinopathy Proliferative, and Macular edema. In this research, author proposes a new strategy for Diabetic Retinopathy can be grouped by combining haar wavelet method and backpropagation. The number of data used were 612 images. The images size 2304x1536, 2240x1536 and 1440x960. The feature extraction of digital image used was haar wavelet at red image, green, and blue at level 1 and level 4 at subband LL and grouping with backpropagation with learning rate 0,1; 0,01 dan 0,001; the division percentage of training data and test data were 70:30, 80:20, 90:10 and 95:5, the value of MSE used was 10-6, epoch maximum 100.000 iteration. The results of this research is the highest test accuracy obtained is 56,25% with image size 2440x1448, haar level 4th and the percentage of comparative training data and test data 95:5, Learning rate 0,1;0,01 and 0,001. Thereby, haar wavelet algorithm cannot identify the feature of diabetic retinopathy and the decomposition process will eliminate much information from diabetic retinopathy