SOCA-YOLO: Smart Optic with Coordinate Attention Model for Vision System-Based Eye Disease Detection

Abstract

Purpose: The purpose of this research is to identify eye diseases using a modified YOLOv9. In particular, we modified YOLOv9 with the addition of Coordinate Attention (CA) for better eye disease detection performance, the use of Programmable Gradient Information (PGI), and Generalized Efficient Layer Aggregation Network (GELAN) for higher computational efficiency and accuracy.

Methods: This study consists of several stages, including the acquisition of eye disease data obtained from the Roboflow website, data annotation, image augmentation, modeling using a modified YOLOv9, and model evaluation.

Result: SOCA-YOLO model achieved an F1 score of 87,2% and mAP50 of 92,9%, outperforming YOLOv9-e by 1,7%. It also surpassed YOLOv6-L6 by 11,1%, YOLOv10-X by 0,8% in mAP50, and YOLOv8-X by 1,1% in recall, showcasing its superior detection accuracy and recall performance.

Novelty: This research contributes by introducing the SOCA-YOLO model in improving the performance of the YOLOv9 by modifying the addition of Coordinate Attention (CA) for better eye disease detection performance, alongside Programmable Gradient Information (PGI) and Generalized Efficient Layer Aggregation Network (GELAN) for better computational efficiency and accuracy.