<![CDATA[Eye disease are visual impairments that can lead to blindness if not detected early. Fundus imaging is one of the most effective methods for identifying abnormalities in the eye. With the advancement of deep neural network technologies, particularly Convolutional Neural Network (CNN), the classification of fundus image can now be performed efficiently. LeNet is a well-known CNN architecture commonly used in image classification tasks, however it has limitation when processing images with complex visual features with high resolution, such as fundus images. This study proposes a modification to the LeNet architecture to enhance it’s a ability to extract important features from images with high resolution. The modification involves adding convolutional layers and adjusting image resolution to optimize the models performance in detecting eye disease in fundus images. The dataset used consists of 4,217 fundus images, classified into four categories: normal, cataract, glaucoma, and diabetic retinopathy. Experimental result show that the original LeNet-5 achieved an accuracy 0f 76%, while the modified LeNet architecture improved the accuracy to 86%. The main contibution of this research lies in the development of a modified and lighweight LeNet architecture, which is capable of handling high-resolution fundus images while maintainig computational efficiency and producing better classification performance compared to the original LeNet.]]>
Copyrights © 2026
