Geographical mapping of coffee crops by using convolutional networks

• Autor(a) principal: Rafael Marlon Peirera Costa Baeta Carreira
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/43132
• topic: Remote sensing; Coffee classification; Convolutional neural networks; Computação – Teses.; Sensoriamento remoto - Teses; Redes neurais convolucionais – Teses; Café – Cultivo – Teses.

In the last decades we have observed a constant growth in the use of remote sensing images for the monitoring of activities and phenomena on Earth allowing the development of several applications. Among the existing applications, the creation of thematic maps is one of the most common, since it allows the classi cation and analysis of the various objects that composes an image and can be used for many purposes, such as: monitoring, planning and recognition. Thematic maps are, usually, generated manually or by the use of models trained by supervised learning. In this type of learning, the system is trained to learn di erent patterns by using labeled samples provided by the user. In this sense, in this dissertation, a thematic map generation method was developed for the recognition of co ee crops in order to obtain data from this crop. For, despite its great importance in the country's economy and Minas Gerais, data collection is still performed manually. The method developed in this work is based on the combination of convolutional neural networks in multiple scales and the choice by neural networks for the development of this project is attributed to the fact that its performance is superior to the traditional methods proposed in computer vision and also not yet be widely used in tasks related to the agricultural area. The use of a multi-scale approach is related to the variation of the size of the patterns found in satellite images and aims to make the method more robust by allowing distinct features to be learned at each of the scales and used in a complementary way.