AI-ready datasets for proof-of-concept weed detection using drones

Document Type

Conference Proceeding

Publication Date

8-2024

Conference Title

23rd Australasian Weeds Conference - Breaking the Cycle: Towards Sustainable Weed Mangement

Place of Publication

Brisbane

ISBN

978-0-646-70156-1

Keywords

AI, weed detection, remote sensing

Disciplines

Artificial Intelligence and Robotics | Weed Science

Abstract

High throughput surveillance is often the key to success of weed eradication programs, particularly for those inconspicuous in the landscape. Although remote sensing would be preferred to address this problem, it can be challenging and costly.

Artificial Intelligence (AI) is rapidly developing and becoming readily accessible, allowing to solve problems previously unsolvable with alternative methods, especially image analysis on diverse conditions, or surrounding environment variability.

This paper presents the process to plan, create, and manage imagery datasets for the detection of two declared species under eradication: skeleton weed (Chondrilla juncea) and rubber vine (Cryptostegia grandiflora) in Western Australia. It is currently implemented employing open-source software for annotating imagery, training detection models, and performing data versioning and experiment tracking.

This low-cost process enables Subject Matter Experts (SMEs) within weed managing agencies to be intrinsically involved in developing high-quality datasets and model prediction results. This feedback process improves model’s predictive output, helping achieve eradication goals and allowing to keep up with the rapid innovations in AI developments.

Link to Full Text

Share

COinS