Computer Vision-Based Deep Learning Algorithms for Detecting Marine Life and Physical Phenomena from Acoustic Backscatter Time Series

点击翻译按钮获取中文摘要

Large quantities of data are constantly acquired during underwater acoustic surveys for environmental monitoring and resources management. The data, visualized as 2D images, are typically analyzed manually or semi-automatically by experts (marine biologists, acousticians, oceanographers), which is time-consuming and prone to errors and inter-expert disagreements. The goal of the proposed research is to develop new software tools for the automated processing and analysis of underwater acoustic data acquired with echosounders, using computer vision-based deep learning methods.We anticipate that this research, carried out in partnership with ASL Environmental Sciences Inc., a British Columbian company, will allow for the automatic detection of marine life, such as eulachon, sandlance, arctic cod, jellyfish, zooplankton, as well as various phenomena near the sea surface and sea bottom, such as air bubbles, waves, ice keels, and suspended sediments, from underwater acoustic data. The potential impacts are significant with respect to efforts in species abundance tracking and environmental monitoring, allowing for a switch from the traditional data analyses towards novel automatic methods reducing processing times, required man-power, and inconsistencies in the results. In addition, this research focuses on regions (depths) that are typically discarded as too difficult to analyze manually. An impact of the automated processing associated with this research is that vast amounts of potentially valuable data may be analyzed rather than discarded, offering a more comprehensive picture of the oceans.The topic is of high importance to Canada as a global leader in sustainable fisheries and ocean resource management, and in research around underwater climate change impacts. The developed software will benefit Canada by providing Canadian researchers with improved and economical means for acoustic data analysis.