Deep learning applied to the classification of corn foliar diseases

Abstract

The search for the development of intelligent models capable of solving complex problems is increasingly common in various fields. One of them is agriculture, where diseases are a major concern, mainly due to the potential loss of productivity and their social and ecological impact. Thus, the use of technology to aid decision-making can be an ally in monitoring crops and, consequently, ensuring successful harvests. Deep Learning is a subfield of Machine Learning that has achieved successful cases in developing intelligent models, particularly in image detection and classification, through the use of Convolutional Neural Networks. In this context, the present work aimed to evaluate Deep Learning-based models for classifying corn leaf diseases by analyzing regions in leaf images. The methodology involved the use of transfer learning, applying the ResNet50 and VGG19 Neural Networks to a subset of publicly available data with 3.838 leaf images, divided into four classes. The results indicated a maximum accuracy of 98,31% using the VGG19 Neural Network and data augmentation techniques.