Drones and machine learning to map sharks in real time

drones detecting sharks in western australia

[Editor’s Note: The Sentinel VDS team visited Shark Bay in October to test their automated shark detection software.]

This post originally appeared on http://www.spatialsource.com.au/unmanned/drones-machine-learning-map-sharks-real-time.

In 2016 there have been 33 reported shark attacks across Australia, two of which were fatal.

Both of these tragic events took place in Western Australia. It is somewhat fitting, therefore, that two key Perth- based consultancies have teamed up to create safer beaches with the latest technology and—crucially—without harming the important and endangered animals themselves.

Environmental consultancy Astron, alongside fellow Perth spatial intelligence and aerial surveying firm C4D Intel, have announced successful completion of their first field trial for their automated shark detection system. Known as Sentinel VDS, the solution uses remotely piloted aircraft systems (RPAS also known as drones or UAV) to map and communicate information about shark movements in real time.

Currently being developed for roll out on Australian and international beaches, Sentinel VDS captures live video, while sophisticated detection software operates autonomously to warn beach users if target shark species are detected.

In October 2016 the Sentinel VDS team visited Shark Bay to trial their automated shark detection software, detecting this 3.5m tiger shark. Video: Astron/C4D Intel

The project was launched earlier this year after Directors from Astron realised the positive impact that RPAS and intelligent detection software, could have on the shark attack problem being experienced in places like Western Australia and northern NSW.

“The use of drones and the development of remote sensing analytics is one of our key services at Astron, so it’s with great optimism that we have launched the Sentinel VDS platform,” explains Astron’s technical director Julian Kruger. “We think that Sentinel VDS will allow the implementation of a shark attack mitigation approach which doesn’t unnecessarily cull sharks and impact other species.”

In early October 2016, the Sentinel VDS team travelled to Shark Bay in Western Australia in order to achieve a significant milestone in the development of the platform.

“Shark Bay seemed an obvious choice as our first trial site with its prevalence of tiger sharks and diversity of habitat and substrate,” said Mr Kruger.

The field campaign allowed the Sentinel VDS team to obtain collection of verified shark imagery which was then used as a training set to improve upon the species detection algorithm.

With the successful completion of first stage trial Mr Kruger noted that a substantial milestone was achieved towards putting the solution to use.

“Our ambition with the Shark Bay field trial was to capture enough data to allow us to firstly confirm, and secondly improve upon, the veracity of the detection software. To have successfully captured the data and then see it produce a positive and persistent detection result was a tremendous outcome for us.”

The team was able to successfully trial a number of operational parameters for the RPAS platform, including flights at varying altitudes, speeds and environmental conditions.

“These results will allow us to fine tune the operational requirements for the UAV platform and optimise the ability of Sentinel VDS to detect sharks, process the imagery and alert the operator within the required operational timeframe,” explains Mr Kruger.

Having confirmed the concept as a valid approach towards shark detection, Astron and C4D Intel will now enter the next stages of development for Sentinel VDS, aiming for operational trials to take place by mid-2017.

State government authorities have granted access that allows Astron and C4D Intel to operate the Sentinel VDS platform in a number designated areas along the Western Australia coast. Therefore, the Sentinel VDS team will be busy over the next six months capturing as much shark footage as they can to inform the machine learning algorithms.

“All extra data from here will help us confirm and improve upon our existing technique,” said Mr Kruger.

“Machine learning is a data intensive approach, so the more training data we can feed it the better it gets.”

FacebookTwitterGoogle+Share

Leave a Reply