Applying Fully Convolutional Architectures for Semantic Segmentation of a Single Tree Species in Urban Environment on High Resolution UAV Optical Imagery

被引：106

作者：

Lobo Torres, Daliana ^{[1
]}

Queiroz Feitosa, Raul ^{[1
]}

Nigri Happ, Patrick ^{[1
]}

Elena Cue La Rosa, Laura ^{[1
]}

Marcato Junior, Jose ^{[2
]}

Martins, Jose ^{[2
]}

Ola Bressan, Patrik ^{[3
,4
]}

Goncalves, Wesley Nunes ^{[2
,4
]}

Liesenberg, Veraldo ^{[5
]}

机构：

[1] Pontifical Catholic Univ Rio De Janeiro, Dept Elect Engn, BR-22451900 Rio De Janeiro, Brazil

[2] Univ Fed Mato Grosso do Sul, Fac Engn Architecture & Urbanism & Geog, BR-79070900 Campo Grande, MS, Brazil

[3] Fed Inst Mato Grosso Sul, BR-79240000 Jardim, Brazil

[4] Univ Fed Mato Grosso do Sul, Fac Comp Sci, BR-79070900 Campo Grande, MS, Brazil

[5] Santa Catarina State Univ, Dept Forest Engn, BR-88520000 Lages, SC, Brazil

来源：

SENSORS | 2020年 / 20卷 / 02期

关键词：

deep learning; fully convolution neural networks; semantic segmentation; unmanned aerial vehicle (UAV); AERIAL VEHICLES; LANDSAT-TM; CLASSIFICATION;

D O I：

10.3390/s20020563

中图分类号：

O65 [分析化学];

学科分类号：

070302 ; 081704 ;

摘要：

This study proposes and evaluates five deep fully convolutional networks (FCNs) for the semantic segmentation of a single tree species: SegNet, U-Net, FC-DenseNet, and two DeepLabv3+ variants. The performance of the FCN designs is evaluated experimentally in terms of classification accuracy and computational load. We also verify the benefits of fully connected conditional random fields (CRFs) as a post-processing step to improve the segmentation maps. The analysis is conducted on a set of images captured by an RGB camera aboard a UAV flying over an urban area. The dataset also contains a mask that indicates the occurrence of an endangered species called Dipteryx alata Vogel, also known as cumbaru, taken as the species to be identified. The experimental analysis shows the effectiveness of each design and reports average overall accuracy ranging from 88.9% to 96.7%, an F1-score between 87.0% and 96.1%, and IoU from 77.1% to 92.5%. We also realize that CRF consistently improves the performance, but at a high computational cost.

引用

页数：20

共 59 条

[1] Urban tree species mapping using hyperspectral and lidar data fusion [J].