Feature selection for tree species identification in very high resolution satellite images

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

Abstract

The aim of this study was to provide an effective feature selection for tree species classifiers in mixed-species boreal forest, from a very high resolution optical satellite image. The 35 input features were the 5 input spectral bands (multispectral and panchromatic channels), 9 contextual features derived from the panchromatic channel and 21 segment-wise features computed at three segment sizes around the treetop locations. A variable ranking was first performed to evaluate the relevance of each feature. Then sequential forward selection was carried out using k-nearest neighbors (kNN) and Linear Discriminant Analysis classifiers. The results suggested that a reasonable feature set would contain 6 to 10 features, mostly from input bands and contextual features. On such a feature set, the best kNN classifier (k=5) returned classification accuracies of 76% for pine and spruce and 88% for decidous trees, with RMS errors between 1.4% and 3.5% and few mixing with the 4 non-tree classes. (6 refs.)
Original languageEnglish
Title of host publicationProceedings
Subtitle of host publicationIEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011
Place of PublicationPiscataway, NJ, USA
PublisherInstitute of Electrical and Electronic Engineers IEEE
Pages4461-4464
ISBN (Electronic)978-1-4577-1005-6
ISBN (Print)978-1-4577-1003-2
DOIs
Publication statusPublished - 2011
MoE publication typeA4 Article in a conference publication
EventIEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011 - Vancouver, BC, Canada
Duration: 24 Jul 201129 Jul 2011

Conference

ConferenceIEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011
Abbreviated titleIGARSS 2011
CountryCanada
CityVancouver, BC
Period24/07/1129/07/11

Fingerprint

discriminant analysis
boreal forest
ranking
satellite image
spectral band

Cite this

Molinier, M., & Astola, H. (2011). Feature selection for tree species identification in very high resolution satellite images. In Proceedings: IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011 (pp. 4461-4464). Piscataway, NJ, USA: Institute of Electrical and Electronic Engineers IEEE. https://doi.org/10.1109/IGARSS.2011.6132538
Molinier, Matthieu ; Astola, Heikki. / Feature selection for tree species identification in very high resolution satellite images. Proceedings: IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011. Piscataway, NJ, USA : Institute of Electrical and Electronic Engineers IEEE, 2011. pp. 4461-4464
@inproceedings{9724a4bc7d414b029a26da9964817412,
title = "Feature selection for tree species identification in very high resolution satellite images",
abstract = "The aim of this study was to provide an effective feature selection for tree species classifiers in mixed-species boreal forest, from a very high resolution optical satellite image. The 35 input features were the 5 input spectral bands (multispectral and panchromatic channels), 9 contextual features derived from the panchromatic channel and 21 segment-wise features computed at three segment sizes around the treetop locations. A variable ranking was first performed to evaluate the relevance of each feature. Then sequential forward selection was carried out using k-nearest neighbors (kNN) and Linear Discriminant Analysis classifiers. The results suggested that a reasonable feature set would contain 6 to 10 features, mostly from input bands and contextual features. On such a feature set, the best kNN classifier (k=5) returned classification accuracies of 76{\%} for pine and spruce and 88{\%} for decidous trees, with RMS errors between 1.4{\%} and 3.5{\%} and few mixing with the 4 non-tree classes. (6 refs.)",
author = "Matthieu Molinier and Heikki Astola",
year = "2011",
doi = "10.1109/IGARSS.2011.6132538",
language = "English",
isbn = "978-1-4577-1003-2",
pages = "4461--4464",
booktitle = "Proceedings",
publisher = "Institute of Electrical and Electronic Engineers IEEE",
address = "United States",

}

Molinier, M & Astola, H 2011, Feature selection for tree species identification in very high resolution satellite images. in Proceedings: IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011. Institute of Electrical and Electronic Engineers IEEE, Piscataway, NJ, USA, pp. 4461-4464, IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011, Vancouver, BC, Canada, 24/07/11. https://doi.org/10.1109/IGARSS.2011.6132538

Feature selection for tree species identification in very high resolution satellite images. / Molinier, Matthieu; Astola, Heikki.

Proceedings: IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011. Piscataway, NJ, USA : Institute of Electrical and Electronic Engineers IEEE, 2011. p. 4461-4464.

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

TY - GEN

T1 - Feature selection for tree species identification in very high resolution satellite images

AU - Molinier, Matthieu

AU - Astola, Heikki

PY - 2011

Y1 - 2011

N2 - The aim of this study was to provide an effective feature selection for tree species classifiers in mixed-species boreal forest, from a very high resolution optical satellite image. The 35 input features were the 5 input spectral bands (multispectral and panchromatic channels), 9 contextual features derived from the panchromatic channel and 21 segment-wise features computed at three segment sizes around the treetop locations. A variable ranking was first performed to evaluate the relevance of each feature. Then sequential forward selection was carried out using k-nearest neighbors (kNN) and Linear Discriminant Analysis classifiers. The results suggested that a reasonable feature set would contain 6 to 10 features, mostly from input bands and contextual features. On such a feature set, the best kNN classifier (k=5) returned classification accuracies of 76% for pine and spruce and 88% for decidous trees, with RMS errors between 1.4% and 3.5% and few mixing with the 4 non-tree classes. (6 refs.)

AB - The aim of this study was to provide an effective feature selection for tree species classifiers in mixed-species boreal forest, from a very high resolution optical satellite image. The 35 input features were the 5 input spectral bands (multispectral and panchromatic channels), 9 contextual features derived from the panchromatic channel and 21 segment-wise features computed at three segment sizes around the treetop locations. A variable ranking was first performed to evaluate the relevance of each feature. Then sequential forward selection was carried out using k-nearest neighbors (kNN) and Linear Discriminant Analysis classifiers. The results suggested that a reasonable feature set would contain 6 to 10 features, mostly from input bands and contextual features. On such a feature set, the best kNN classifier (k=5) returned classification accuracies of 76% for pine and spruce and 88% for decidous trees, with RMS errors between 1.4% and 3.5% and few mixing with the 4 non-tree classes. (6 refs.)

U2 - 10.1109/IGARSS.2011.6132538

DO - 10.1109/IGARSS.2011.6132538

M3 - Conference article in proceedings

SN - 978-1-4577-1003-2

SP - 4461

EP - 4464

BT - Proceedings

PB - Institute of Electrical and Electronic Engineers IEEE

CY - Piscataway, NJ, USA

ER -

Molinier M, Astola H. Feature selection for tree species identification in very high resolution satellite images. In Proceedings: IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2011. Piscataway, NJ, USA: Institute of Electrical and Electronic Engineers IEEE. 2011. p. 4461-4464 https://doi.org/10.1109/IGARSS.2011.6132538