An Approach for Disease and Pest Detection in Onion Crops using MobileNetV3-S, EfficientNetV2-S, and ResNet50

Acuña D.; Melgar A.; Wong L. · Applications in Software Engineering - Proceedings of the 14th International Conference on Software Process Improvement, CIMPS 2025 (2025) · DOI: 10.1109/cimps69356.2025.11419683

Resumen

In Peruvian agriculture, onion crops suffer significant losses due to the emergence of diseases and pests, a situation worsened by the lack of accessible technological tools for early detection. Traditional visual identification proves insufficient, especially in rural environments with limited resources. The purpose of this study is to evaluate the performance of three deep learning models, including EfficientNetV2-S, MobileNetV3-S, and ResNet50, for the detection of diseases and pests in onion crops. The methodology was structured into four stages: image acquisition from the TOM2024 dataset, application of data augmentation techniques to balance the classes, development and training of three deep learning models in a controlled environment, and performance evaluation through cross-validation. The results show that EfficientNetV2-S achieved an average precision of ninety-eight percent and an accuracy of ninety-six percent, outperforming the other architectures. These findings highlight the potential of this model to improve agricultural diagnostics.

Instituciones: Universidad Peruana de Ciencias Aplicadas