Aplicação de métodos de processamento de imagem e redes neurais na extração de características de doenças em plantas

Abstract

Currently, one of the biggest problems in agriculture is related to disease control, being responsible for great losses. The identification of diseases is a crucial factor for the control of pathologies, so it is important to make a quick decision. This research investigated three methods of extracting characteristics for disease classification, namely, color histograms, texture histogram and artificial neural networks. Different sets of images were used, two sets of cotton images, one from the Digipathos repository and the PlantVillage. The OpenCV and Scikitlearn libraries were used to obtain the histograms and the TensorFlow and Keras libraries were used to develop the neural networks. The VGG network model was also used to assess the transfer of learning. The sets of images were separated into training, validation and testing, in which the results were evaluated with the set of tests. The best performances were obtained, first with the neural networks followed by the HSV color histogram, LBP texture histogram and RGB color histogram.