Automatic training sample selection for a multi-evidence based crop classification approach

Menka Chellasamy, Paul A Ferre, Mogens Humlekrog Greeve

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

An approach to use the available agricultural parcel information to automatically select training samples for crop classification is investigated. Previous research addressed the multi-evidence crop classification approach using an ensemble classifier. This first produced confidence measures using three Multi-Layer Perceptron (MLP) neural networks trained separately with spectral, texture and vegetation indices; classification labels were then assigned based on Endorsement Theory. The present study proposes an approach to feed this ensemble classifier with automatically selected training samples. The available vector data representing crop boundaries with corresponding crop codes are used as a source for training samples. These vector data are created by farmers to support subsidy claims and are, therefore, prone to errors such as mislabeling of crop codes and boundary digitization errors. The proposed approach is named as ECRA (Ensemble based Cluster Refinement Approach). ECRA first automatically removes mislabeled samples and then selects the refined training samples in an iterative training-reclassification scheme. Mislabel removal is based on the expectation that mislabels in each class will be far from cluster centroid. However, this must be a soft constraint, especially when working with a hypothesis space that does not contain a good approximation of the targets classes. Difficulty in finding a good approximation often exists either due to less informative data or a large hypothesis space. Thus this approach uses the spectral, texture and indices domains in an ensemble framework to iteratively remove the mislabeled pixels from the crop clusters declared by the farmers. Once the clusters are refined, the selected border samples are used for final learning and the unknown samples are classified using the multi-evidence approach. The study is implemented with WorldView-2 multispectral imagery acquired for a study area containing 10 crop classes. The proposed approach is compared with the multi-evidence approach based on training samples selected randomly and border samples based on initial cluster centroids within agricultural parcels without any refinement. The results clarify the improvement in overall classification accuracy to 82.3% based on the proposed approach from 74.9 % based on random selection and 71.4% on non-refined border samples.

Original languageEnglish (US)
Title of host publicationInternational Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives
PublisherInternational Society for Photogrammetry and Remote Sensing
Pages63-69
Number of pages7
Volume40
Edition7
DOIs
StatePublished - 2014
EventISPRS Technical Commission VII Mid-Term Symposium 2014 - Istanbul, Turkey
Duration: Sep 29 2014Oct 2 2014

Other

OtherISPRS Technical Commission VII Mid-Term Symposium 2014
CountryTurkey
CityIstanbul
Period9/29/1410/2/14

Fingerprint

Crops
crop
evidence
Classifiers
Textures
Analog to digital conversion
farmer
Multilayer neural networks
texture
Labels
digitization
Pixels
worldview
Neural networks
neural network
vegetation index
subsidy
confidence
pixel
imagery

Keywords

  • Classification
  • Crop
  • Imagery
  • Learning
  • Neural
  • Satellite
  • Training
  • Vector

ASJC Scopus subject areas

  • Information Systems
  • Geography, Planning and Development

Cite this

Chellasamy, M., Ferre, P. A., & Greeve, M. H. (2014). Automatic training sample selection for a multi-evidence based crop classification approach. In International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives (7 ed., Vol. 40, pp. 63-69). International Society for Photogrammetry and Remote Sensing. https://doi.org/10.5194/isprsarchives-XL-7-63-2014

Automatic training sample selection for a multi-evidence based crop classification approach. / Chellasamy, Menka; Ferre, Paul A; Greeve, Mogens Humlekrog.

International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives. Vol. 40 7. ed. International Society for Photogrammetry and Remote Sensing, 2014. p. 63-69.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Chellasamy, M, Ferre, PA & Greeve, MH 2014, Automatic training sample selection for a multi-evidence based crop classification approach. in International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives. 7 edn, vol. 40, International Society for Photogrammetry and Remote Sensing, pp. 63-69, ISPRS Technical Commission VII Mid-Term Symposium 2014, Istanbul, Turkey, 9/29/14. https://doi.org/10.5194/isprsarchives-XL-7-63-2014
Chellasamy M, Ferre PA, Greeve MH. Automatic training sample selection for a multi-evidence based crop classification approach. In International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives. 7 ed. Vol. 40. International Society for Photogrammetry and Remote Sensing. 2014. p. 63-69 https://doi.org/10.5194/isprsarchives-XL-7-63-2014
Chellasamy, Menka ; Ferre, Paul A ; Greeve, Mogens Humlekrog. / Automatic training sample selection for a multi-evidence based crop classification approach. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives. Vol. 40 7. ed. International Society for Photogrammetry and Remote Sensing, 2014. pp. 63-69
@inproceedings{79ca7223c2d140b199bfc99925cd1a63,
title = "Automatic training sample selection for a multi-evidence based crop classification approach",
abstract = "An approach to use the available agricultural parcel information to automatically select training samples for crop classification is investigated. Previous research addressed the multi-evidence crop classification approach using an ensemble classifier. This first produced confidence measures using three Multi-Layer Perceptron (MLP) neural networks trained separately with spectral, texture and vegetation indices; classification labels were then assigned based on Endorsement Theory. The present study proposes an approach to feed this ensemble classifier with automatically selected training samples. The available vector data representing crop boundaries with corresponding crop codes are used as a source for training samples. These vector data are created by farmers to support subsidy claims and are, therefore, prone to errors such as mislabeling of crop codes and boundary digitization errors. The proposed approach is named as ECRA (Ensemble based Cluster Refinement Approach). ECRA first automatically removes mislabeled samples and then selects the refined training samples in an iterative training-reclassification scheme. Mislabel removal is based on the expectation that mislabels in each class will be far from cluster centroid. However, this must be a soft constraint, especially when working with a hypothesis space that does not contain a good approximation of the targets classes. Difficulty in finding a good approximation often exists either due to less informative data or a large hypothesis space. Thus this approach uses the spectral, texture and indices domains in an ensemble framework to iteratively remove the mislabeled pixels from the crop clusters declared by the farmers. Once the clusters are refined, the selected border samples are used for final learning and the unknown samples are classified using the multi-evidence approach. The study is implemented with WorldView-2 multispectral imagery acquired for a study area containing 10 crop classes. The proposed approach is compared with the multi-evidence approach based on training samples selected randomly and border samples based on initial cluster centroids within agricultural parcels without any refinement. The results clarify the improvement in overall classification accuracy to 82.3{\%} based on the proposed approach from 74.9 {\%} based on random selection and 71.4{\%} on non-refined border samples.",
keywords = "Classification, Crop, Imagery, Learning, Neural, Satellite, Training, Vector",
author = "Menka Chellasamy and Ferre, {Paul A} and Greeve, {Mogens Humlekrog}",
year = "2014",
doi = "10.5194/isprsarchives-XL-7-63-2014",
language = "English (US)",
volume = "40",
pages = "63--69",
booktitle = "International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives",
publisher = "International Society for Photogrammetry and Remote Sensing",
edition = "7",

}

TY - GEN

T1 - Automatic training sample selection for a multi-evidence based crop classification approach

AU - Chellasamy, Menka

AU - Ferre, Paul A

AU - Greeve, Mogens Humlekrog

PY - 2014

Y1 - 2014

N2 - An approach to use the available agricultural parcel information to automatically select training samples for crop classification is investigated. Previous research addressed the multi-evidence crop classification approach using an ensemble classifier. This first produced confidence measures using three Multi-Layer Perceptron (MLP) neural networks trained separately with spectral, texture and vegetation indices; classification labels were then assigned based on Endorsement Theory. The present study proposes an approach to feed this ensemble classifier with automatically selected training samples. The available vector data representing crop boundaries with corresponding crop codes are used as a source for training samples. These vector data are created by farmers to support subsidy claims and are, therefore, prone to errors such as mislabeling of crop codes and boundary digitization errors. The proposed approach is named as ECRA (Ensemble based Cluster Refinement Approach). ECRA first automatically removes mislabeled samples and then selects the refined training samples in an iterative training-reclassification scheme. Mislabel removal is based on the expectation that mislabels in each class will be far from cluster centroid. However, this must be a soft constraint, especially when working with a hypothesis space that does not contain a good approximation of the targets classes. Difficulty in finding a good approximation often exists either due to less informative data or a large hypothesis space. Thus this approach uses the spectral, texture and indices domains in an ensemble framework to iteratively remove the mislabeled pixels from the crop clusters declared by the farmers. Once the clusters are refined, the selected border samples are used for final learning and the unknown samples are classified using the multi-evidence approach. The study is implemented with WorldView-2 multispectral imagery acquired for a study area containing 10 crop classes. The proposed approach is compared with the multi-evidence approach based on training samples selected randomly and border samples based on initial cluster centroids within agricultural parcels without any refinement. The results clarify the improvement in overall classification accuracy to 82.3% based on the proposed approach from 74.9 % based on random selection and 71.4% on non-refined border samples.

AB - An approach to use the available agricultural parcel information to automatically select training samples for crop classification is investigated. Previous research addressed the multi-evidence crop classification approach using an ensemble classifier. This first produced confidence measures using three Multi-Layer Perceptron (MLP) neural networks trained separately with spectral, texture and vegetation indices; classification labels were then assigned based on Endorsement Theory. The present study proposes an approach to feed this ensemble classifier with automatically selected training samples. The available vector data representing crop boundaries with corresponding crop codes are used as a source for training samples. These vector data are created by farmers to support subsidy claims and are, therefore, prone to errors such as mislabeling of crop codes and boundary digitization errors. The proposed approach is named as ECRA (Ensemble based Cluster Refinement Approach). ECRA first automatically removes mislabeled samples and then selects the refined training samples in an iterative training-reclassification scheme. Mislabel removal is based on the expectation that mislabels in each class will be far from cluster centroid. However, this must be a soft constraint, especially when working with a hypothesis space that does not contain a good approximation of the targets classes. Difficulty in finding a good approximation often exists either due to less informative data or a large hypothesis space. Thus this approach uses the spectral, texture and indices domains in an ensemble framework to iteratively remove the mislabeled pixels from the crop clusters declared by the farmers. Once the clusters are refined, the selected border samples are used for final learning and the unknown samples are classified using the multi-evidence approach. The study is implemented with WorldView-2 multispectral imagery acquired for a study area containing 10 crop classes. The proposed approach is compared with the multi-evidence approach based on training samples selected randomly and border samples based on initial cluster centroids within agricultural parcels without any refinement. The results clarify the improvement in overall classification accuracy to 82.3% based on the proposed approach from 74.9 % based on random selection and 71.4% on non-refined border samples.

KW - Classification

KW - Crop

KW - Imagery

KW - Learning

KW - Neural

KW - Satellite

KW - Training

KW - Vector

UR - http://www.scopus.com/inward/record.url?scp=84918537832&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84918537832&partnerID=8YFLogxK

U2 - 10.5194/isprsarchives-XL-7-63-2014

DO - 10.5194/isprsarchives-XL-7-63-2014

M3 - Conference contribution

VL - 40

SP - 63

EP - 69

BT - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives

PB - International Society for Photogrammetry and Remote Sensing

ER -