GlorotUniform Kernel Initializer based Conv2D Sequential Convolutional Neural Network for Diabetic Retinopathy Detection

People with diabetes may develop Diabetic Retinopathy (DR), an sight illness that damages the eye nerve vessels in the retina and may end up in vision loss and blurred vision. Retinal camera expenses, time restrictions, an absence of expertise in DR inspection, made it challenging to implement DR evaluation as a general practice, which led to the use of modern technology towards the detection of chronic ailments regardless of a firm faith in its effectiveness of controlling them. This paper offers suggestions by designing GlorotUniform Kernel Initializer Conv2D (GKIC) Sequential model for addressing this problem by experimenting the Indian DR Image Dataset. The Indian DR Dataset from KAGGLE, which contains 3363 retinal tissue scan images, has been implemented in the suggested GKIC model. The images are preprocessed and fitted with GKIC model for the detection of DR. The retina scan images are preprocessed with two convolutional layers accompanied max pooling layer followed by two dense layer that are weight initialized with GlorotUniform Kernel initializer to detect the existence of DR. The image files are broken down into training and evaluation images, and the training retinal scan images have been utilized to assess the efficiency of GKIC model and the prevailing conV2D paradigms. Implementation observations using Python on a Gtx Geforce Tesla V100 Nvidia GPU workstation using 64 and 250 training intervals reveals that the proposed GKIC model estimates accuracy of 98.67% when evaluated against existing CNN models.