Cost-effective surveillance of emerging aquatic weeds using robotic aircraft

Summary

In the 2007/2008 Defeating the Weed Menace R&D program, a novel approach to the detection and eradication of emerging aquatic weeds was presented. The Australian Centre for Field Robotics (ACFR) at the University of Sydney proposed the development of an autonomous weed controller: one that wouldn’t tire, would travel large distances, and wouldn’t mind traversing through difficult to access areas in the hope of detecting and eradicating nasty weeds – and so it was, the Weedinator was born.

The ACFR are world renowned for their exploitations in autonomous systems – basically robots that can do all the “dull, dirty and dangerous” tasks that are out there. The group has worked on novel systems, delivered a number of world firsts, and commercialised many of their products in the areas of logistics, mining, defence, agriculture and art.

Aims

Projective objectives were three fold:

1. To provide weed control authorities a cost-effective surveillance tool for the detection of emerging aquatic weeds. The system is an autonomous Hovering Unmanned Air Vehicle (HUAV) – a robotic helicopter – fitted with low-cost sensor suites and intelligent aquatic weed detection algorithms. Such a system can provide coverage over large distances and inaccessible aquatic habitats.
2. To develop a spray mechanism for the HUAV that provides the weed control operators the capability of autonomous or remote control spraying and treatment of these aquatic weeds in inaccessible aquatic habitats.
3. To gather geo-referenced spectral and visual data of aquatic weeds in inaccessible aquatic habitats. This data can be used to conduct further research in spatiotemporal weed growth.

Outcomes

The platform is a modified G18 model helicopter from UVA Vision www.uavision.com . Using a helicopter meant full manoeuvrability could be attained, including hover, thus giving the ability to traverse large distances, move into tight situations, and hold position to take imagery or to spray. This involved development and tuning of flight control and navigation algorithms, as well the spray mechanisms. The final system could fly for approximately two hours and carry approximately 500ml of herbicide (although water was used in this project for demonstration).

The detection algorithms were based on novel machine learning techniques that are currently very popular in the scientific community. The basic principle is that instead of developing algorithms that specifically try to model what a weed is (this approach has always failed in the past), why not develop an algorithm that learns the model itself. Such approaches mimic the way the human brain learns.  With such an algorithm you give it many images of the weed of interest, and many images that are not the weed of interest. You tell the algorithm which is what, and then leave it alone while it learns the key differences, focussing on qualities such as colour, shape and texture.

The algorithms proved to be very robust and accurate and were tested on Alligator weed and sprayed Salvinia.

Background

Waterways often include places that are difficult to access.  When aquatic weeds establish in these areas they present problems both for detection of the weeds and for treatment to manage their spread or eradicate them.

Using small unmanned aircraft equipped with low-cost sensing cameras and combining these with new technologies to analyse spectral data obtained, different plant species can be ‘finger-printed’ in ways that allow weeds to be detected in places that were previously relatively inaccessible.

Publications and Resources

• Australian Centre for Field Robotics

Citation

Land & Water Australia. 2008. Cost-effective surveillance of emerging aquatic weeds using robotic aircraft. [Online] (Updated May 1st, 2009)
Available at: http://lwa.gov.au/node/2584 [Accessed Wednesday 29th of February 2012 08:07:48 PM ].