Deteccão de ruas de barro a partir de imagens aéreas

Abstract

This study focuses on the development of an arti cial intelligence system in the Deep Learning area, using image processing techniques to build the desired model, with the objective of detecting unpaved roads with an aerial view of cities using convolutional neural networks . The primary objective is to o er a tool capable of providing information that can be used to monitor cities, such as tracking the growth of urban and rural areas, road paving planning and to apply public policies, such as basic sanitation and security. The semantic segmentation technique and the U-Net architecture were used to build the proposed model, with an IoU of 0.7188, metric used to evaluate the proposed model. The results showed that the network was able to correctly identify most of the dirt roads present in the dataset, but there is still room for improvement. A larger database and better hardware resources can help improve the performance of the developed model.