Predicción de plagas en cultivos de arroz utilizando Redes Neuronales Convolucionales

Abstract

Introduction: Agriculture in the Sucre region faces severe economic losses due to late detection of pests in rice crops. Timely pest detection in agricultural crops is critical to ensuring productivity and food security. In Sucre, Colombia, rice crops are essential to the regional economy. However, the low level of technological development in the field makes it difficult to implement efficient pest monitoring solutions. This article presents the development of an accessible platform that integrates a convolutional neural network (CNN) model to predict pests with high accuracy.

Objective: To develop and validate a computer vision-based pest prediction system using a convolutional neural network trained from scratch.

Methodology: Images of rice crops affected by various pests were collected. A CNN model was designed from scratch, trained in Google Colab, and integrated into a platform developed in FastAPI for consumption through a web interface. Results: The model achieved an accuracy of over 85% in classification tests, with an average response time per image of less than five seconds.

Conclusion: The proposed solution enables early, accessible, and efficient detection for farmers in Sucre, contributing to food security and crop sustainability.