Desarrollo de una herramienta computacional basada en redes neuronales para el diagnóstico del tizón tardío en cultivos de papa

Abstract

Late blight (Phythottora Infestas) is a disease that seriously affects potato crops, causing a negative impact on the farmer's economy. This project will generate a computational tool called NeuroPI - 2105 based on a convolutional neural network created by the author, which classifies two types of leafs that are; healthy and sick. The network has been trained with 1304 images from the PlantVillage database, 304 of them correspond to healthy leaves, the remainder being attributed to late blight. The training algorithm has used the Adam gradient descent, the cross-entropy error function, and backpropagation, in order to adjust the synaptic weights and threshold levels in the network.