A machine learning approach to crop localisation using spatial information

Authors:
Blair Howarth;Jayantha Katupitiya;Ray Eaton;Sarath Kodagoda
Affiliations:
School of Mechanical and Manufacturing Engineering, The University of New South Wales, Kensington, Sydney, Australia.;School of Mechanical and Manufacturing Engineering, The University of New South Wales, Kensington, Sydney, Australia.;School of Electrical Engineering and Telecommunications, The University of New South Wales, Kensington, Sydney, Australia.;Mechatronics and Intelligent Systems Group, Faculty of Engineering, University of Technology Sydney Broadway, NSW 2007, Australia
Venue:
International Journal of Computer Applications in Technology
Year:
2010

Citing 1
Cited 0

An Agricultural Mobile Robot with Vision-Based Perception for Mechanical Weed Control

Autonomous Robots

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper describes an approach to recognise and localise centres of mature lettuce heads in the field when the lettuce leaves obscure the distinctions between plants. This is of great value when using an automatic harvester in cluttered or closely planted vegetation. The aim of this work is to investigate and verify the potential use of spatial rather than visual clues for recognition and localisation, with a view to implement a more robust and sophisticated system if promise is shown. Colour/texture information was difficult to use so spatial information was used instead. A laser range finder was used to generate a height plot from above the plants. Lettuce examples were used to learn the radial distribution of the lettuce model. This was compared with the distributions of arbitrary locations in new scans to locate possible lettuce locations. Planting distance information was then used to localise the final lettuce positions. The algorithm was able to successfully locate 15 out of 16 sample lettuces.