High Speed Crop and Weed Identification in Lettuce Fields for Precision Weeding.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Plant Science

Background:

• Herbicides pose environmental risks, necessitating sustainable weed management strategies.
• Precision weeding requires accurate, high-speed identification of individual plants.
• Current methods often lack the speed or cost-effectiveness for real-time field application.

Purpose of the Study:

• To develop and validate a system for high-precision, low-cost identification of crops and weeds in lettuce fields.
• To enable herbicide reduction or elimination through individual plant targeting.
• To achieve accurate plant localization for robotic weeding systems.

Main Methods:

• Utilized red, green, and near-infrared (NIR) reflectance with size differentiation for plant identification.
• Employed LED illumination at 525, 650, and 850 nm with a modified RGB camera for single-shot image capture.
• Implemented a kinematic stereo method to correct parallax errors and determine precise plant locations.

Main Results:

• In-field weed and crop identification rates were 56% and 69%, respectively, at speeds up to 10 km/h.
• Post-trial processing significantly improved identification accuracy to 81% for weeds and 88% for crops.
• The system demonstrated effectiveness across various lettuce growth stages.

Conclusions:

• The developed system shows significant potential for enabling precision weeding and reducing herbicide reliance.
• High-speed, accurate plant identification is feasible using multispectral imaging and kinematic stereo.
• Further optimization of post-trial processing can enhance real-time weed detection capabilities.