東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

FindBook

Google Book

Amazon

博客來

Forest Conservation with Deep Learning: A Deeper Understanding of the Human Geography Around the Betampona Nature Reserve, Madagascar.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Forest Conservation with Deep Learning: A Deeper Understanding of the Human Geography Around the Betampona Nature Reserve, Madagascar./
作者:	Cota, Gizelle Rebecca.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	70 p.
附註:	Source: Masters Abstracts International, Volume: 83-05.
Contained By:	Masters Abstracts International83-05.
標題:	Remote sensing. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28717337
ISBN:	9798480645897

Forest Conservation with Deep Learning: A Deeper Understanding of the Human Geography Around the Betampona Nature Reserve, Madagascar.
Cota, Gizelle Rebecca.

Forest Conservation with Deep Learning: A Deeper Understanding of the Human Geography Around the Betampona Nature Reserve, Madagascar. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 70 p.

Source: Masters Abstracts International, Volume: 83-05.

Thesis (M.Sc.)--Saint Louis University, 2021.

This item must not be sold to any third party vendors.

Documenting the coupled impact of climate change and human activities on tropical rainforest ecosystems is imperative, not only for protecting tropical biodiversity but also for better implementation of U.N. Sustainable Development Goals. Recent advances in very high-resolution satellite sensor systems (i.e., WorldView-3), computing power, and machine learning has provided improved mapping of fine scale changes in the tropics. However, approaches so far have been focused on feature extraction or pixel-based methods that require extensive manual tuning of machine learning parameters hindering the potential of machine learning in forest conservation mapping, not taking advantage of texture information that are found to be powerful for many applications. Additionally, the contribution of Shortwave Infrared (SWIR) bands in forest habitat mapping is unknown. The objectives of this study were to develop end-to-end, automated mapping of tropical forest habitat using fully convolution neural networks (FCNNs) with WorldView-3 (WV-3) imagery and evaluate human impact on the tropical environment using the Betampona Nature Reserve (BNR) in Madagascar as the test site. An FCNN model (U-Net) that can utilize spatial/textural information was implemented, and other feature-fed pixel-based machine learning methods including Support Vector Machine (SVM), Random Forest (RF), and Deep Neural Network (DNN) were used for comparison. Our results show that FCNN model outperformed other pixel-based models with an accuracy of 90.9%, while SVM, RF and DNN provided accuracies of 88.6%, 84.8%, and 86.6%, respectively. (2) When SWIR bands were excluded from the input data, FCNN approach still provided superior performance over other methods with a negligible decrease (by 1.87%) in accuracies while accuracies of other pixel-based models SVM, RF, and DNN decreased significantly by 5.42%, 3.18%, and 8.55%, respectively. (3) Spatial-temporal analysis of land-cover and land-use types showed 0.7% increase in the Evergreen Forest class within BNR and 32% increase in tree cover within residential areas surrounding BNR from 2010 to 2019, likely due to forest regeneration and conservation efforts. The paper also discusses the effect of conservation efforts such as distribution of fuel-efficient stoves to mitigate firewood harvesting for cooking and agroforestry to improve food security.

ISBN: 9798480645897Subjects--Topical Terms:

535394
Remote sensing.
Subjects--Index Terms:

Conservation efforts`

Forest Conservation with Deep Learning: A Deeper Understanding of the Human Geography Around the Betampona Nature Reserve, Madagascar.
LDR:03597nmm a2200385 4500 001 2350306
005 20221020125235.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798480645897
035 $a (MiAaPQ)AAI28717337
035 $a AAI28717337
040 $a MiAaPQ $c MiAaPQ
100 1 $a Cota, Gizelle Rebecca. $3 3689778
245 1 0 $a Forest Conservation with Deep Learning: A Deeper Understanding of the Human Geography Around the Betampona Nature Reserve, Madagascar.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 70 p.
500 $a Source: Masters Abstracts International, Volume: 83-05.
500 $a Advisor: Sagan, Vasit.
502 $a Thesis (M.Sc.)--Saint Louis University, 2021.
506 $a This item must not be sold to any third party vendors.
520 $a Documenting the coupled impact of climate change and human activities on tropical rainforest ecosystems is imperative, not only for protecting tropical biodiversity but also for better implementation of U.N. Sustainable Development Goals. Recent advances in very high-resolution satellite sensor systems (i.e., WorldView-3), computing power, and machine learning has provided improved mapping of fine scale changes in the tropics. However, approaches so far have been focused on feature extraction or pixel-based methods that require extensive manual tuning of machine learning parameters hindering the potential of machine learning in forest conservation mapping, not taking advantage of texture information that are found to be powerful for many applications. Additionally, the contribution of Shortwave Infrared (SWIR) bands in forest habitat mapping is unknown. The objectives of this study were to develop end-to-end, automated mapping of tropical forest habitat using fully convolution neural networks (FCNNs) with WorldView-3 (WV-3) imagery and evaluate human impact on the tropical environment using the Betampona Nature Reserve (BNR) in Madagascar as the test site. An FCNN model (U-Net) that can utilize spatial/textural information was implemented, and other feature-fed pixel-based machine learning methods including Support Vector Machine (SVM), Random Forest (RF), and Deep Neural Network (DNN) were used for comparison. Our results show that FCNN model outperformed other pixel-based models with an accuracy of 90.9%, while SVM, RF and DNN provided accuracies of 88.6%, 84.8%, and 86.6%, respectively. (2) When SWIR bands were excluded from the input data, FCNN approach still provided superior performance over other methods with a negligible decrease (by 1.87%) in accuracies while accuracies of other pixel-based models SVM, RF, and DNN decreased significantly by 5.42%, 3.18%, and 8.55%, respectively. (3) Spatial-temporal analysis of land-cover and land-use types showed 0.7% increase in the Evergreen Forest class within BNR and 32% increase in tree cover within residential areas surrounding BNR from 2010 to 2019, likely due to forest regeneration and conservation efforts. The paper also discusses the effect of conservation efforts such as distribution of fuel-efficient stoves to mitigate firewood harvesting for cooking and agroforestry to improve food security.
590 $a School code: 0193.
650 4 $a Remote sensing. $3 535394
650 4 $a Artificial intelligence. $3 516317
650 4 $a Conservation biology. $3 535736
653 $a Conservation efforts`
653 $a FCN
653 $a Forest cover mapping
653 $a Land cover mapping
653 $a SWIR
653 $a WorldView-3
690 $a 0799
690 $a 0800
690 $a 0408
710 2 $a Saint Louis University. $b Geoscience. $3 1677473
773 0 $t Masters Abstracts International $g 83-05.
790 $a 0193
791 $a M.Sc.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28717337