This project has developed an underwater video recording system with machine-learning technology to monitor and quantify lobster presence near fishing activities. The machine aims to ultimately be used to inform and advise decision-makers, fishers and researchers investigating fishery ecology.
The project addresses the need for a tool to assess trends in the abundance of commercially targeted lobster stocks. Owing to the cryptic nature of the species, and the complexity of its behavioural response to fishing gear, this is fundamentally a very difficult task. In this project, we have developed a low-cost underwater video system mounted on a standard creel (lobster trap) frame that can be baited and deployed alongside commercial fishing gear to capture images of lobsters on the fishing grounds. This is then coupled with machine-learning systems trained to identify the presence of lobsters in the images, aimed at automating the process of generating an index of lobster abundance. In this feasibility study we have successfully demonstrated the operation of a low-cost creel-mounted video system, deployed at sea by a commercial fisher. Use of underwater video footage of lobsters to train a deep convolutional neural network has successfully demonstrated the concept of automated detection of lobsters. Ultimately the combined package of creel-mounted camera system and automated image processing will provide a tool of value to fishers, fishery managers and research scientists.