Stereovision system and image processing algorithms for plant specific application

Hong Y. Jeon, Lei Tian, Tony Grift, Loren Bode, Aaron Hager

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

Abstract

A stereovision system with image processing algorithm was developed to identify weeds and estimate their positions from the field image. The developed algorithm was designed to automatically segment plants against soil background under multiple illuminations, sunny and shady, in the viewing area. The algorithm segmented plants against soil background and identified individual plants in field images with the processing errors of 2.9 % for the stationary images and 2.1 % for the images captured while the vision system was moving. The algorithm estimated three-dimensional (3D) coordinates of plants in images and a reasonable range of the application inaccuracy was expected due to the limitation in sensing resolution and disparity matching of the stereovision. Root mean square (RMS) error of the estimation was from 43.1 mm to 71.8 mm in 3D coordinates, and the correlation between the measured and estimated 3D coordinates reduces the RMS error to 7.1-12.2 mm. Artificial Neural Network (ANN) was used to identify weeds among the detected plants. Normalized patterns of plants were supplied to the ANN for training and validating the network, respectively. The ANN identified 72.6 % of corn plants in the field images. The main sources of the identification error were identified and additional identification criteria were applied to improve the identification rate: the improvement of the identification rates, 92.5 % and 95.1 % were noted. The stereovision system with developed algorithm may provide a unique technique in sensing plants against the soil background in the image under variable outdoor illuminations without any manual process. The developed system may be useful for autonomous machine-vision based field applications i.e. field navigation, plant specific weed control, plant population mapping and field scouting.

Original languageEnglish (US)
Title of host publicationAmerican Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009
Pages147-170
Number of pages24
StatePublished - Dec 1 2009
EventAmerican Society of Agricultural and Biological Engineers Annual International Meeting 2009 - Reno, NV, United States
Duration: Jun 21 2009Jun 24 2009

Publication series

NameAmerican Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009
Volume1

Other

OtherAmerican Society of Agricultural and Biological Engineers Annual International Meeting 2009
CountryUnited States
CityReno, NV
Period6/21/096/24/09

Fingerprint

image analysis
neural networks
Soil
Lighting
lighting
weeds
soil
Weed Control
computer vision
weed control
Zea mays
Research Design
Population

Keywords

  • Image processing
  • Image segmentation
  • Machine vision
  • Stereovision

ASJC Scopus subject areas

  • Agricultural and Biological Sciences(all)

Cite this

Jeon, H. Y., Tian, L., Grift, T., Bode, L., & Hager, A. (2009). Stereovision system and image processing algorithms for plant specific application. In American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009 (pp. 147-170). (American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009; Vol. 1).

Stereovision system and image processing algorithms for plant specific application. / Jeon, Hong Y.; Tian, Lei; Grift, Tony; Bode, Loren; Hager, Aaron.

American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009. 2009. p. 147-170 (American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009; Vol. 1).

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

Jeon, HY, Tian, L, Grift, T, Bode, L & Hager, A 2009, Stereovision system and image processing algorithms for plant specific application. in American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009. American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009, vol. 1, pp. 147-170, American Society of Agricultural and Biological Engineers Annual International Meeting 2009, Reno, NV, United States, 6/21/09.
Jeon HY, Tian L, Grift T, Bode L, Hager A. Stereovision system and image processing algorithms for plant specific application. In American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009. 2009. p. 147-170. (American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009).
Jeon, Hong Y. ; Tian, Lei ; Grift, Tony ; Bode, Loren ; Hager, Aaron. / Stereovision system and image processing algorithms for plant specific application. American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009. 2009. pp. 147-170 (American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009).
@inproceedings{19785ea962584dda8ae2528505d73b00,
title = "Stereovision system and image processing algorithms for plant specific application",
abstract = "A stereovision system with image processing algorithm was developed to identify weeds and estimate their positions from the field image. The developed algorithm was designed to automatically segment plants against soil background under multiple illuminations, sunny and shady, in the viewing area. The algorithm segmented plants against soil background and identified individual plants in field images with the processing errors of 2.9 {\%} for the stationary images and 2.1 {\%} for the images captured while the vision system was moving. The algorithm estimated three-dimensional (3D) coordinates of plants in images and a reasonable range of the application inaccuracy was expected due to the limitation in sensing resolution and disparity matching of the stereovision. Root mean square (RMS) error of the estimation was from 43.1 mm to 71.8 mm in 3D coordinates, and the correlation between the measured and estimated 3D coordinates reduces the RMS error to 7.1-12.2 mm. Artificial Neural Network (ANN) was used to identify weeds among the detected plants. Normalized patterns of plants were supplied to the ANN for training and validating the network, respectively. The ANN identified 72.6 {\%} of corn plants in the field images. The main sources of the identification error were identified and additional identification criteria were applied to improve the identification rate: the improvement of the identification rates, 92.5 {\%} and 95.1 {\%} were noted. The stereovision system with developed algorithm may provide a unique technique in sensing plants against the soil background in the image under variable outdoor illuminations without any manual process. The developed system may be useful for autonomous machine-vision based field applications i.e. field navigation, plant specific weed control, plant population mapping and field scouting.",
keywords = "Image processing, Image segmentation, Machine vision, Stereovision",
author = "Jeon, {Hong Y.} and Lei Tian and Tony Grift and Loren Bode and Aaron Hager",
year = "2009",
month = "12",
day = "1",
language = "English (US)",
isbn = "9781615673629",
series = "American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009",
pages = "147--170",
booktitle = "American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009",

}

TY - GEN

T1 - Stereovision system and image processing algorithms for plant specific application

AU - Jeon, Hong Y.

AU - Tian, Lei

AU - Grift, Tony

AU - Bode, Loren

AU - Hager, Aaron

PY - 2009/12/1

Y1 - 2009/12/1

N2 - A stereovision system with image processing algorithm was developed to identify weeds and estimate their positions from the field image. The developed algorithm was designed to automatically segment plants against soil background under multiple illuminations, sunny and shady, in the viewing area. The algorithm segmented plants against soil background and identified individual plants in field images with the processing errors of 2.9 % for the stationary images and 2.1 % for the images captured while the vision system was moving. The algorithm estimated three-dimensional (3D) coordinates of plants in images and a reasonable range of the application inaccuracy was expected due to the limitation in sensing resolution and disparity matching of the stereovision. Root mean square (RMS) error of the estimation was from 43.1 mm to 71.8 mm in 3D coordinates, and the correlation between the measured and estimated 3D coordinates reduces the RMS error to 7.1-12.2 mm. Artificial Neural Network (ANN) was used to identify weeds among the detected plants. Normalized patterns of plants were supplied to the ANN for training and validating the network, respectively. The ANN identified 72.6 % of corn plants in the field images. The main sources of the identification error were identified and additional identification criteria were applied to improve the identification rate: the improvement of the identification rates, 92.5 % and 95.1 % were noted. The stereovision system with developed algorithm may provide a unique technique in sensing plants against the soil background in the image under variable outdoor illuminations without any manual process. The developed system may be useful for autonomous machine-vision based field applications i.e. field navigation, plant specific weed control, plant population mapping and field scouting.

AB - A stereovision system with image processing algorithm was developed to identify weeds and estimate their positions from the field image. The developed algorithm was designed to automatically segment plants against soil background under multiple illuminations, sunny and shady, in the viewing area. The algorithm segmented plants against soil background and identified individual plants in field images with the processing errors of 2.9 % for the stationary images and 2.1 % for the images captured while the vision system was moving. The algorithm estimated three-dimensional (3D) coordinates of plants in images and a reasonable range of the application inaccuracy was expected due to the limitation in sensing resolution and disparity matching of the stereovision. Root mean square (RMS) error of the estimation was from 43.1 mm to 71.8 mm in 3D coordinates, and the correlation between the measured and estimated 3D coordinates reduces the RMS error to 7.1-12.2 mm. Artificial Neural Network (ANN) was used to identify weeds among the detected plants. Normalized patterns of plants were supplied to the ANN for training and validating the network, respectively. The ANN identified 72.6 % of corn plants in the field images. The main sources of the identification error were identified and additional identification criteria were applied to improve the identification rate: the improvement of the identification rates, 92.5 % and 95.1 % were noted. The stereovision system with developed algorithm may provide a unique technique in sensing plants against the soil background in the image under variable outdoor illuminations without any manual process. The developed system may be useful for autonomous machine-vision based field applications i.e. field navigation, plant specific weed control, plant population mapping and field scouting.

KW - Image processing

KW - Image segmentation

KW - Machine vision

KW - Stereovision

UR - http://www.scopus.com/inward/record.url?scp=76549099342&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=76549099342&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:76549099342

SN - 9781615673629

T3 - American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009

SP - 147

EP - 170

BT - American Society of Agricultural and Biological Engineers Annual International Meeting 2009, ASABE 2009

ER -