Deep Learning making waves in underwater applications

Founded in Seattle, USA, (2017) MarineSitu is a company that specialises in the development of integrated machine vision cameras for monitoring marine life and underwater environments. MarineSitu's initial focus was on creating technology to monitor conditions around devices like wave energy converters and underwater turbines that are used in the emerging renewable marine energy sector known as the "blue economy." The blue economy relates to sustainable development of marine resources.

President and Founder, James Joslin says that MarineSitu's systems are being used for different underwater applications. An example is installations that track and counts fish at several hydroelectric facilities, which will help inform fisheries management of fish populations migrating up rivers.

 

Most cameras that perform underwater applications, often mounted on underwater drones or ROV's, are limited by the duration they can stay deployed. After just a few months submerged, underwater systems are required to be brought up to the surface for cleaning and maintenance. Joslin says they were interested in designing a system that would be capable of spending years deployed underwater due to the nature of the application.

"Monitoring marine energy devices requires long-term continuous monitoring in order to catch potentially rare events, such as marine mammal interactions with tidal turbines," Joslin says.

The primary target for their technology was the blue energy industry. To achieve this, MarineSitu designed a system that comprises two main components: hardware and software. The hardware includes Machine Vision Cameras from Allied Vision, ruggedized housings, lighting systems, and controllers. The cameras used are Allied Vision's Alvium GigE cameras (5Gigabit bandwidth, 34.1MP, and 464 FPS) that are fitted into specialized housings designed for prolonged underwater applications. These housings are mainly constructed with plastic as they must be resistant to corrosion. They also incorporate Kowa LM5JCM lenses and include features like mechanical wipers to maintain lens clarity.

 

When asked, James Joslin answered "We chose these cameras and lenses after doing a review of machine vision systems for a combination of factors including the high level of configuration control that they offer, their low-light sensitivity, the broad range of resolution and frame rate, as well as their compact size."

In the gloomy depths lighting and visibility are key challenges. Another reason why MarineSitu chose the Alvium Camera, Joslin said, was due to its I/O capabilities and ability to synchronize with up to eight LED lights per camera which are strobed. The low LED light duty cycle in this design minimises the effect of bright lighting on the surrounding marine environment.

"Our hardware is specifically developed for long-term deployments and has been demonstrated with over 4 years in the water. Compared to other similar hardware on the market it is also significantly lower cost," Joslin says.

The camera equipment and hardware that MarineSitu have elected to use has proved to be very successful and able to withstand long-periods of deployment underwater. A system in Oahu, Hawaii, has now been in-place and operating successfully for over 20-months.

MarineSitu's system digitizes data collected by the Alvium camera via an ethernet connection over fibre, with a fibre optic cable connected to a land-based data processor. Ethernet Fibre optic cabling provides reliable and fast transmission over long distances.

Having solved the issues associated with capturing and collecting high quality machine vision data in underwater environments MarineSitu turned their attention to the second part of the system, being software and the task of monitoring the environment. The project began with people working around the clock monitoring a computer screen in six-hour shifts counting fish. Using human-power to monitor 24/7 was extremely labour intensive, resource draining, and presented the need for an automated software solution.

Underwater target detection is complicated by the highly dynamic marine environment so trained AI models help identify the targets of interest. To develop the automated software solution, MarineSitu partnered with AI solutions developer Plainsight. Together they successfully developed AI models trained to efficiently parse large volumes of image data from extended underwater footage and so reduce the volume of data that needed to be viewed by human-operators by 95%.

Joslin says that MarineSitu has now implemented its vision system across various locations in the United States and has plans for growth in the immediate future.

For more information about this system, Allied Vision Cameras, Kowa Lenses or advice on how to design a Machine Vision system contact Adept Turnkey's machine vision experts.