Development of Deep Learning Based Application System for the Classification of Farm Related Ocular Trauma in Benue State

Abstract

Farm related ocular diseases constitute a major public health challenge among agricultural workers, particularly in developing regions where access to specialized eye care is limited. Ocular diseases such as cataract, glaucoma, and retinopathy are prevalent among farmers due to prolonged exposure to sunlight, dust, chemicals, and poor occupational safety practices. Despite the growing burden of these conditions, limited studies have explored the application of artificial intelligence–based diagnostic systems using localized data from Benue State, Nigeria. This study presents the development of a deep learning–based application system for the classification of farm related ocular diseases in Benue State. A total of 2,715 ocular images were collected from 85 subjects diagnosed with cataract, glaucoma, and retinopathy at Okida Eye Clinic, Otukpo. The dataset was augmented to improve class balance and diversity, and transfer learning was applied using a pre-trained AlexNet model with frozen convolutional layers. Model performance was evaluated using accuracy and loss metrics during training in a Python environment. The proposed model achieved a classification accuracy of 96.5% with loss values below 0.2, demonstrating strong learning capability and generalization. Comparative analysis with existing state-of-the-art models shows that the proposed approach performs competitively while benefiting from localized clinical data. The trained model was integrated into a software application and tested with real ocular images, yielding high confidence classification scores. The system is therefore recommended as a reliable decision-support tool for early detection and management of farm related ocular diseases in Benue State.