Uma rede neural convolucional para segmentação de imagens de minério de ferro

Abstract

This work presents a mechanism used to search a Convolutional Neural Network (CNN) based on the architecture of a network called UNet. The main aim is to segment Ąne iron ore regions in an image to facilitate the iron ore production process in a plant. The images were collected from an iron ore plant in a screening step. It was possible to obtain a dataset composed of 688 images and their label segmentation. The results of the optimized architecture shows that the found architecture of CNN achieved value of 80% of Intersect Over Union(IoU) and the other comparison networks: UNet without optimization and DeepLabV3+, which achieved 75% and 78%, respectively. In general, the results found were considered promising and the optimized CNN solution was able to detect the Ąne ore regions