Current weed identification relies on expert knowledge, interpreting taxonomic or morphological identification keys, trying to match a weed with an image in a weed guide or app, or sending a weed image or specimen to herbariums for identification. This often leads to delayed or incorrect identification that impedes timely action, particularly on emerging weeds, and requires State herbaria to allocate scarce resources to respond to community identification requests for well-known established weeds.
This project will develop, trial and implement Australia’s first real-time, artificial intelligence-based, automated identification of national, state and regional priority weeds. It will also develop and promote a fit for purpose community weed management, alert, reporting and communication system – WeedScan – building on the currently available FeralScan platform to better enable cooperative community-led weeds management.
This new tool will provide state and national governments, NRM groups, graziers, farmers and communities with an easy to use digital tool enabling:
(a) Weeds to be identified quickly without expert knowledge.
(b) Easy access to best practice management information.
(c) Enable action either at the individual enterprise level or as part of a cooperative regional WeedScan community-led management and communication system.
The objective of this project is to produce a weed identification app using computer vision and a reporting and information system WeedScan, with the former leading the user to the specific weed information and reporting interface provided by the latter.
This project receives funding from the Australian Government National Landcare Programme
February 2021 update:
CISS/CSIRO have been collecting and collating photos of weeds donated by the public for the purpose of creating training datasets for the weed identification mobile app. Approximately 12,000 labelled photos are now available for training the artificial intelligence identification system.