Deep learning is the most technology in 21century, it gives more information about how computers can understand data and learning from. In deep learning, networks of artificial neurons analyse large dataset to automatically discover patterns. In this paper, we will introduce the part of these techniques to know how we can use deep learning to create our own model to diagnosis eye diseases. The most idea will be addressed is the evaluation performance model using confusion matrix. In this study, we will compare three models of neural network, CNN, Vgg16 and Inceptionv3 in order to evaluate performance of the models. In 0ur work, a deep learning convolutional network based on keras and tensorflow is deployed using python for image classification. a number of different images, which contains four types of eye diseases, namely Diabetic retinopathy, Glaucoma, Myopia and Normal are used for image classification. Three different structures of neural network, CNN, VGG16 and Inception V3 are compared on GPU system in Google Colab, with three different combinations of classifiers. It is shown that, the results for each combination and observed that for multi-image classification, Inception V3 combination gives better classification accuracy (81.00 %) than any other models. Using of confusion matrix showing us where our classifier is confused when it makes prediction.

Keywords : Inception V3, CNN, Vgg16, Eye Diseases, Confusing Matrix, Deep Learning, Diabetic Retinopathy, Glaucoma, Myopia.