Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/138867
Title: Reconocimiento de edificios en imágenes aéreas mediante redes neuronales
Author: González García-Tizón, Francisco de Asís
Tutor: Burguera Burguera, Antonio
Others: Ventura, Carles  
Abstract: The segmentation of aerial images to recognize buildings in an automated way can have a multitude of practical applications in different fields. The present study exemplifies how this task could be carried out empirically. In this TFM we start from a limited dataset in which it is known for each photo which pixels correspond to buildings. As a first step, different techniques were developed to expand the starting dataset until it is practically unlimited, discarding those that are not significant. Subsequently, different models of convolutional neural network models were configured and parameterized, which, taking these data as input, were trained and evaluated to achieved predictions with enough quality about the buildings hidden inside unknown images. Finally, the results were compared with each other. It was observed that, although sequential models capable of obtaining satisfactory results were achieved, in the current work they were beaten by pre-trained models. We also obtained measures of the accuracy of the predictions and visualizations of both the hit and misses made by the network on a batch of example images. The conclusion is that a well-trained network can fight with a human being doing this job and that the development of this technology is of undoubted scientific and social interest. The aim of this project is not to provide a general solution to the problem, but to prove that neural networks are very capable of solving it successfully and that the study can be extended to real cases.
Keywords: aerial images
deep learning
computer vision
Document type: info:eu-repo/semantics/masterThesis
Issue Date: 24-Dec-2021
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
PAC3.ipynb2,76 MBUnknownView/Open
currotizonTFM1221memoria.pdfMemoria del TFM2,21 MBAdobe PDFThumbnail
View/Open