WeedRemeed

Effective weed detection at scale using colour-picking and AI/ML

Australia’s first drone image weed detection solution on the AWS cloud

WeedRemeed – How it works in 5 easy steps

Step 1
Fly and capture

Using a connected drone, the operator scans the appropriate area capturing aerial coverage of the field

Step 2
Make a picking swatch

An intermediate operator then uses a colour picker tool to isolate the colour associated with the target weeds in a sample image

Step 3
Process at scale

The chosen colour is then processed using A.I. and M.L. in the cloud to identify the target weeds among the entire image set taken from the drone operator

Step 4
GeoReference all weeds

The system then identifies all target species among the entire image data set

Step 5
Follow the map – find all weeds easily

With the geo-targeted information provided by WeedRemeed, council operators can then begin the remediation process

See real world examples below

Links to the Australian Scalable Drone Cloud

Solving the ‘big data’ challenge

The WeedRemeed Portal has been developed over 3 years as a regional partnership between Snowy Monaro Regional Council, Local Land Services and 2pi Software.

Cloud-hosted in Australia, WeedRemeed uniquely facilitates secure sharing of large drone footage datasets to meet the inordinately high data needs of Machine Learning and Artificial Intelligence (ML/AI) for weed detection.

The WeedRemeed future roadmap includes a robust security and governance model, and links to third-party services to put real-world machine learning capabilities in the hands of biosecurity professionals, Landcare groups & farmers to remediate priority national weeds and boost farm yields.

The innate scalability and adaptability of the system allows significant ease of set-up and adoption by multiple councils within the near region, and nationally.

Developed through a regional partnership with

Read how we helped the Snowy Monaro Council identify and target African Lovegrass for remediation

Read the case study

Features

Resilient upload (low bandwidth protection)
Temporary cloud links for large file sharing
Easy Colour picking using simple swatches (easily created in MS Paint or other desktop tool)
Multi swatch – search for many colour, shadow and other variations simultaneously
Extensive security measures

Does not require complex Data Science Capabilities to operate
Simultaneous batch runs for faster throughput
Exif extraction
GeoJSON export – QGIS-ready
Image splitting
Links to ASDC and ODM
Data governance options

Headless (API access) for batch or high scale operation
Low cost cloud storage options
Docker friendly integration
Proven techniques for packaging custom processes/transforms: supporting Python, ts/js, Golang, DockerContainers
Available as a bureau service for organisations without in-house capabilities.

The process in depth

Capture

Drone operator(s) in the field use one or multiple drones to capture an assessment area. Assessment areas can be:

large blocks of land (council and BioSecurity applications)
farms (agricultural applications)
parks (land management and BioDiversity research applications)

Bulk data uploads

Resulting images are then uploaded into the WeedRemeed portal. This process can be initiated in the field for on-the-fly processing.

Upload view

In the upload view, the WeedRemeed portal offers the following features:

Secure user access/logins
The ability to set naming conventions for batch uploads
The ability to capture meta data from imported images

Specialist Data Handling

With the image and data imported, WeedRemeed can then execute the following processes:

Swatch creation (multiple)
EXIF extraction
GeoReferencing data generation
Data sampling
Comparisons

The WeedRemeed system allows expert data analysts to fine-tune controls for data to ensure the highest quality data is being analysed and processed.

Common Data Pre-Processing Ops (Mix n Match)

WeedRemeed provides users with the following pre-processing tools:

Splice
Segmentation
Greyscaling
Other operations coming

AI/ML Inference

Using cloud-based AI/ML tools, all of this data is then processed at scale. Data can be used to feed training algorithms that can then be used for future data sets. Reports and other visualisation tools can be generated from all of the resulting data processing.