Comparison of SOM and CNN Models for Automated Disease Diagnosis in Banana Leaves (#255)

Authors

Navarro Tantalean, Daniel Hernan

Vegas Villar, Fernando Isai

Huarote Zegarra, Raul Eduardo

Abstract

Modern agriculture has to face a large number of problems due to the increase of crop diseases, affecting both productivity and economic lines of social groups in rural markets. This paper proposes a mobile application, based on the use of self-organizing neural networks (SOM, Self-Organizing Maps), for the automatic diagnosis of banana leaf diseases. The application allows capturing images from a mobile device (Android), processing them using image processing techniques and classifying them without the need for large volumes of labeled data. Unlike the other approaches such as convolutional neural networks (CNNs), the SOM method reduces computational demands, making it ideal for resource-constrained areas. The model was trained and validated using a database with real images of banana leaves affected by diseases such as Sigatoka, Cordana or Pestalotiosis. The results obtained show a high accuracy of the system, thus validating the effectiveness of the proposed approach for practical, sustainable and low-cost agricultural conditions.