Automating the post-processing of noisy hydroacoustic fish surveying for monitoring tidal turbines

DeepSense partnered with the Fundy Ocean Research Centre for Energy (FORCE) and OERA to create a machine learning model to automate the post-processing and analysis of hydroacoustic fish surveys in extremely noisy environments suitable for tidal turbines such as the Bay of Fundy. FORCE use hydroacoustic echosounder surveys to evaluate the impact of tidal turbines on marine life in the Bay of Fundy.  In order to analyze the survey data, manual pre-processing is currently required to annotate the data. Manual processes are time consuming and can create the opportunity for inconsistency. For this project, a machine learning model was trained to detect and filter noise in the hydroacoustic sensor data. By identifying “bad regions” in the hydroacoustic data with a model and automatically extracting these regions from survey data, the team is able to more accurately and consistently clean data while significantly reducing the manual effort required to pre-process the data.