Image capture and efficacy of weed control for robotic weeding

Robotic weeding: you will help in a project distinguishing weeds from the crop by machine vision and assess the potential efficacy of weed control by droplet applications to individual leaves in the field.

Department: Crop Production

Supervised by: Alistair Murdoch

The Placement Project

Policy issues and legislation are driving the need for a paradigm shift in weed control. Robots offer an engineering as a paradigm shift from the conventional approach of selective chemistry, and even the biotech one of genetically-modifed, herbicide tolerant (GM-HT) crops. This UROP project will be part of the eyeSpot robotic weeding project (AHDB-funded, 2015-2018) with the aim: no chemicals to the crop; no chemicals to the soil; instead leaf-specific droplet applications only to unwanted plants (aka weeds!). A 95-99% reduction in herbicide inputs is predicted compared to GM-HT and conventional methods while enhancing food quality and security. Expertise in more precise farming (PF) in the Crop Production Division of SAPD comes from the 5.25 year eyeWeed project (Innovate-UK, 2010-2015). Meanwhile, the School of Systems Engineering have been investigating the potential of Micro Air Vehicles (MAVs) for remote sensing. The student will carry out field work with others. Images will be captured by eyeWeed system and by MAV. The student will assess “ground” and “image” truth, identifying leaves of weeds and crops. Images will be captured from vegetable plots at Sonning Farm each week. Droplet applications of herbicide will be made to leaves efficacy assessed visually on the ground and in images. Correlations of ground truth, image truth and machine vision algorithm output will ultimately be performed to assess ability of a robot to distinguish wanted (crop) and unwanted plants. So data analysis will relate to ground truth and image truth and treatment efficacy.

Tasks

1. Work with PhD student identifying weeds in vegetable crops at Sonning Farm and on grower's fields 2. Assist in image capture using "eyeWeed" machine vision system mounted on sprayer and also by MAV 3. Examine geo-referenced images using custom-written software and tag weed and crop leaves, identifying species were possible. 4. Compare image truth and ground truth (correlation) 5. Compare image and ground truth data before and after droplet treatments

Skills, knowledge and experience required

1. Basic awareness of anatomy of plants and relevance to weed identification 2. IT skills (and willingness to learn new packages) 3. Understanding of why there is a need for a paradigm shift in approaches to weed control. 4. Willingness to participate actively in field research at forefront of weed science. 5. Willingness to follow health and safety rules for work on farms.

Skills which will be developed during the placement

1. Weed identification 2. Data analysis 3. Experience of novel approaches to weed control which are at the forefront of weed science and addressing major concerns of growers about loss of herbicides and 1. Team working 4. Experimental design 5. Selection of appropriate statistical analysis methods 6. Scientific writing skills in writing up results (potential for conference paper)

Place of Work

University of Reading, Agriculture Building, Harborne Building, Sonning Farm and visits to vegetable growers' fields on some days. Off-site working days may be longer and may involve an overnight stay (accompanied by supervisor)

Hours of Work

30 hours per week (variable)

Approximate Start and End Dates (not fixed)

Monday 13 June 2016 - Friday 29 July 2016

How to Apply

CV and covering letter.Informal interview with PI, Professor Ferryman, Mr Paul de la Warr and PhD student