Estudio y comparativa de algoritmos de detección de objetos con redes neuronales artificiales convolucionales para la detección de enfermedades en hojas

Recently Machine Learning and computational vision have generated interest and have found new applications in engineering. In agriculture, "smart" systems have become important tools in the detection of anomalies that decrease the quality and quantity in the harvest of agricultural products. In this research, we developed a comparison of the main object detection algorithms using Convolutional Neural Networks (CNN) implemented in Deep Learning. The results were analyzed based on the accuracy and processing time obtained with the object detection algorithms R-CNN, Fast R-CNN, and Faster R-CNN. The CCN topologies of AlexNet, GoogleNet, ResNet50, ResNet101, SqueezeNet and InceptionV3 were implemented to generate Transfer Learning in image detectors and classifiers. The topologies were trained with the PlantVillage - Dataset which is made up of more than 40,000 leaf images of 9 plant species and 24 diseases