Improving in-row weed detection in multispectral stereoscopic images

  • Authors:

  • A. Piron;V. Leemans;F. Lebeau;M. -F. Destain

  • Affiliations:

  • Gembloux Agricultural University, Unité de Mécanique et Construction, 2 Passage des Déportés, 5030 Gembloux, Belgium;Gembloux Agricultural University, Unité de Mécanique et Construction, 2 Passage des Déportés, 5030 Gembloux, Belgium;Gembloux Agricultural University, Unité de Mécanique et Construction, 2 Passage des Déportés, 5030 Gembloux, Belgium;Gembloux Agricultural University, Unité de Mécanique et Construction, 2 Passage des Déportés, 5030 Gembloux, Belgium

  • Venue:
  • Computers and Electronics in Agriculture
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Previous research has shown that plant height and spectral reflectance are relevant features to classify crop and weeds in organic carrots: classification based on height gave a classification accuracy (CA) of up to 83% while classification based on a combination of three multispectral bands gave a CA of 72%. The first goal of this study was to examine the simultaneous use of both height and multispectral parameters. It was found that classification rate was only slightly improved when using a feature set comprising both height and multispectral data (2%). The second goal of this study was to improve the detection method based on plant height by setting an automatic threshold between crop and weeds heights, in their early growth stage. This threshold was based on crop row determination and peak detection in plant height probability density function, corresponding to the homogeneous crop population. Using this method, the CA was 82% while the CA obtained with optimal plant height limits is only slightly higher at 86%.