Deep learning-based convolutional neural network structured new image classi cation approach for eye disease identi cation

Abstract

This study implemented a new image classi cation method approach based on a convolutional arti cial neural network using deep learning techniques. Sample application was carried out using the proposed method for the diagnosis of diabetic retinopathy. A problem that may arise here is how to collect information about the blood vessels and identify any abnormal patterns from the rest of the phonoscopic images, and then assess the degree of retinopathy. To solve this problem, the current research proposed a developed methodology and the algorithmic structure of this new approach. An approach called “care model" was utilized in this study, which di ers from the classical Convolutional Neural Network (CNN) structure. The care approach is based on the idea of generating a better solution by incorporating new data obtained through rescaling the available data based on the total number of pixels before creating the average data pool. This allows for the continuation of CNN processes with enhanced accuracy and performance. In this approach, all data sets are multiplied by the number of elements by the number of epoch times eight tensors. The purposed care model included VGG19 image classi cation model and developed a mathematical model. The pre-trained model and all the image dataset were taken from the kaggle and keras to be implemented as the case study. The proposed model provided the training accuracy of 87%, testing accuracy of 88%, precision of 93%, and recall of 83%.