Semantic Segmentation for Pest Detection on Organic Banana Leaves

García P.; Machacuay J.; Manrique J.; Ipanaqué W. · Proceedings - IEEE CHILEAN Conference on Electrical, Electronics Engineering, Information and Communication Technologies, ChileCon (2025) · DOI: 10.1109/chilecon66915.2025.11476279

Resumen

Pest recognition in agriculture currently relies on manual inspections. This approach is subjective, time-consuming, and difficult to scale. In precision agriculture, research studies address this issue for different types of plants by leveraging machine learning models to identify pests in classification and segmentation settings. However, to the best of our knowledge, there exists little or no research on banana leaf pest disease segmentation. To address this gap, this research work develops two segmentation models: one based on the U-Net architecture and the other on the SegFormer architecture, both of which are well known in the literature. The best model, U-Net, achieved an average Dice coefficient of 0.9532. It outperformed SegFormer in accuracy but required higher computational costs. These results and the visual outputs confirm the viability of deep learning for accurate and automated banana pest segmentation.

Instituciones: Universidad de Piura