Over 800 species of coral provide a habitat for around a quarter of the world’s marine life, but they are being degraded by over fishing, bottom trawling, warming temperatures and unsustainable coastal development.
According to Ewen Plougastel, managing director and ASEAN delivery lead, Accenture Applied Intelligence, a video camera equipped with AI is working in depths between five and 10 metres to give marine biologists at the Sulubaaï Environmental Foundation (SEF) a better understanding of the subsea environment.
Since 2019 Accenture, Intel and SEF have established Project: CORaiL, which has included the installation of an artificial, concrete reef dubbed the Sulu-Reef Prosthesis (SRP). The SRP was placed in the reef surrounding Pangatalan Island before fragments of living coral were planted on it to grow, expand and provide a hybrid habitat for fish and marine life.
Underwater video cameras - equipped with the Accenture Applied Intelligence Video Analytics Services Platform (VASP) - detect and photograph fish as they pass by. VASP uses AI, powered by Intel Xeon, Intel FPGA Programmable Acceleration Cards and Intel Movidius VPU, to count and classify the marine life. 4G technology is used to send data through a wireless router to shore, Plougastel said via email.
“The first version of the software deployed on the camera is used to purely collect images of fish to create a dataset that we could use to train a deep learning model that would then be capable of classifying marine life automatically. This version is equipped with classic computer vision algorithms detecting movements and saving an image of the “object” - here the fish or other marine species - that passed by the camera," said Plougastel.
“The second version of the software will be automatically detecting and classifying the fish in real-time. The detections will then be sent to a centralised server with more processing power that will run further analytics to derive more details about the marine life species.”
Marine biologists at the Sulubaaï foundation move the camera once a week to get a variety of angles and locations on the reef. Since being deployed in May 2019, it has collected roughly 40,000 images.
The camera contains a very sensitive sensor to capture images during periods of low-light during the day. A second version of the camera is currently in development that could include infra-red capabilities for footage at night.
“We would ideally use either a depth camera or a stereoscopic camera to also derive the size of the marine life in front of the camera, which is another important datapoint for the marine biologists,” Plougastel said.