Author(s): Shrestra, Sanjeevan
Date: 2018
Persistent ID: http://hdl.handle.net/10362/33652
Origin: Repositório Institucional da UNL
Subject(s): Building Extraction; High Resolution Aerial Imagery; Deep Learning; Deep Convolutional Neural Network; Fully Convolutional Network; Conditional Random Field
Dissertation submitted in partial fulfilment of the requirements for the degree of Master of Science in Geospatial Technologies
Building extraction from remotely sensed imagery plays an important role in urban planning, disaster management, navigation, updating geographic databases and several other geospatial applications. Several published contributions are dedicated to the applications of Deep Convolutional Neural Network (DCNN) for building extraction using aerial/satellite imagery exists; however, in all these contributions a good accuracy is always paid at the price of extremely complex and large network architectures. In this paper, we present an enhanced Fully Convolutional Network (FCN) framework especially molded for building extraction of remotely sensed images by applying Conditional Random Field (CRF). The main purpose here is to propose a framework which balances maximum accuracy with less network complexity. The modern activation function called Exponential Linear Unit (ELU) is applied to improve the performance of the Fully Convolutional Network (FCN), resulting in more, yet accurate building prediction. To further reduce the noise (false classified buildings) and to sharpen the boundary of the buildings, a post processing CRF is added at the end of the adopted Convolutional Neural Network (CNN) framework. The experiments were conducted on Massachusetts building aerial imagery. The results show that our proposed framework outperformed FCN baseline, which is the existing baseline framework for semantic segmentation, in term of performance measure, the F1-score and Intersection Over Union (IoU) measure. Additionally, the proposed method stood superior to the pre-existing classifier for building extraction using the same dataset in terms of performance measure and network complexity at once.
