Diseño de un detector del hongo moniliasis en cultivos de cacao para pequeños productores

Abstract

Currently the cocoa sector in Ecuador occupies large areas, in the province of Sto. Domingo de los Tsáchilas, Valle Hermoso parish, there are a large number of small cocoa producers, for whom the detection of the Moniliasis Fungus is highly complex, however, they carry out an efficient control manually to eliminate the fungus. For this work, three phases of the fungus are defined that vary by the percentage of the size, the stain and the appearance of the gray powder in the cocoa pod. The present work has as objective the design of a detector of the moniliasis fungus in cocoa crops for small producers through the use of Matlab software which includes image processing techniques, the different methods that can be used for the detection of images are investigated. , opting for the Deep Learning option which allows you to create basic convolutional neural networks and use already created convolutional networks that can be adapted to a new task by transferring learning. Several algorithms are created and adapted to severely detect the black spot with gray dust that is caused by the Moniliasis Fungus in the cocoa pod. For the development of these algorithms, three convolutional neural networks are chosen, which are: Basic Neural Network that is designed from scratch, GoogLeNet Convolutional Neural Network and ResNet-18 Convolutional Neural Network, these last two networks are available in the Matlab database. Using the Matlab App Designer tool, a user-friendly interface is created, in which an image of the cocoa pod is selected and it proceeds to define the phase in which the fungus is found.