Publicação

Informal settlement segmentation using VHR RGB and height information from UAV imagery: a case study of Nepal

Ver documento

Detalhes bibliográficos
Resumo:Informal settlement in developing countries are complex. They are contextually and radiometrically very similar to formal settlement. Resolution offered by Remote sensing is not sufficient to capture high variations and feature size in informal settlements in these situations. UAV imageries offers solution with higher resolution. Incorporating UAV image and normalized DSM obtained from UAV provides an opportunity of including information on 3D space. This can be a crucial factor for informal settlement extraction in countries like Nepal. While formal and informal settlements have similar texture, they differ significantly in height. In this regard, we propose segmentation of informal settlement of Nepal using UAV and normalized DSM, against traditional approach of orthophoto only or orthophoto and DSM. Absolute height, normalized DSM(nDSM) and vegetation index from visual band added to 8 bit RGB channels are used to locate informal settlements. Segmentation including nDSM resulted in 6 % increment in Intersection over Union for informal settlements. IoU of 85% for informal settlement is obtained using nDSM trained end to end on Resnet18 based Unet. Use of threshold value had same effect as using absolute height, meaning use of threshold does not alter result from using absolute nDSM. Integration of height as additional band showed better performance over model that trained height separately. Interestingly, benefits of vegetation index is limited to settlements with small huts partly covered with vegetation, which has no or negative effect elsewhere.
Autores principais:Sigdel, Ganesh Prasad
Assunto:Informal Settlement Unmanned Aerial Vehicle (UAV) Deep Learning Semantic Segmentation Normalized Digital Surface Model(nDSM) Visible-Band Difference Vegetation Index (VDVI)
Ano:2021
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL

Registos relacionados

Registos relacionados