JelliGuard

The Problem

Jellyfish are one of the world’s most invasive marine species. From habitat intrusion to massive swarms, Jellyfish are responsible for the destruction of ecosystems, shutting down everything from beaches to nuclear power plants, and significant economic loss for the marine and tourism industries. Invasive jellyfish are often carried in ballast water to new regions and once there they wreak havoc on the local ecology. Jellyfish blooms have shut down major industrial infrastructure including nuclear and desalination plants in the USA, Sweden and Japan. In the Mediterranean alone, jellyfish sting more than 150,000 people a year, resulting in beach closures that affect tourism across the region. The situation is getting worse, jellyfish are one of the few marines species that actually thrive when humans mess up. Overfishing, pollution, and climate change cause more favourable conditions for jellyfish and allow them to dominate species that are negatively impacted. Increased numbers of jellyfish in specific regions therefore point to underlying man-made problems in the marine environment. However, you’ve got to love these little guys. They are able to invade and take over the world without a single brain cell between them. Jellyfish do not have brains and act purely on impulse. Though don’t underestimated them. The tracking, mapping and forecasting of jellyfish concentrations is vital to protect ecosystems, beachgoers, and the fishing industry.

Our Proposal

JelliGuard is a network of underwater vision sensors that use trained AI image processing to identify jellyfish. The sensors lie on the seabed and their cameras point up through the water column to detect jellyfish as they pass. Standard camera trap triggers are not suitable underwater so we will use pure image processing. However, image recognition can be computationally intensive requiring cloud solutions or high power, not really suitable for battery devices. The unique feature of our underwater cameras will be their use of two levels of image processing to achieve low-power, long-term operation while still maintaining a high detection rate. The camera’s monitor mode will use a low resolution, low frame rate, and computationally simple image algorithm that is optimized for the lowest possible power-consumption. In this mode we target an image matching accuracy of 60% to 70%. When the monitor mode makes an identification, the system will switch to a high-power mode to take high-resolution images and use a more computationally intensive algorithm to achieve a high accuracy and confirm the detection. The sensor data will be used to monitor jellyfish concentrations and provide an early-warning system. Current jellyfish countermeasures cannot be economically deployed across large areas so it is vital to know where and when they can be effectively used. JelliGuard will provide detection in near real-time and will guide the efficient deployment of jellyfish defences.

We Assume that...

1. A small number of vision sensors can provide adequate coverage to detect jellyfish invasion over a larger area.

2. Detection can then be used to trigger action and effective jellyfish countermeasures can be deployed based on the sensor data.

3. We can collect, or find, enough sample images to train our image recognition algorithm accurately.

4. We can meet the low-power targets and the vision sensors can operate underwater for significant periods of time.

Constraints to Overcome

AI image recognition algorithms that can be used on battery-powered microprocessors in embedded systems. A way of triggering camera traps based on specific species.

Current Work

Create a prototype vision sensor that can be used to test our assumptions and demonstrate it can accurately detect jellyfish at the required level of power consumption.

Current Needs

People with experience in machine recognition algorithms especially for embedded systems. Anyone who has used camera traps in a marine environment and is willing to provide feedback on features and use. Marketing experts who can drive a concept through to commercial launch. Potential end-users in jelly infested waters who could help trial and test the prototype.