Tracking the health of underwater species is critical to understanding the effects of climate change on marine ecosystems. Unfortunately, it’s a time-consuming process — biologists conduct studies with echosounders that use sonar to determine water and object depth, and they manually interpret the resulting 2D echograms. These interpretations are often prone to error and require pricey software like Echoview.
Fortunately, a team of research scientists hailing from the University of Victoria in Canada are developing a machine learning method for detecting specific biological targets in acoustic survey data. In a preprint paper (“A Deep Learning based Framework for the Detection of Schools of Herring in Echograms“), they say that their approach — which they tested on schools of herring — might measurably improve the accuracy of environmental monitoring.
The team trained an AI model on a corpus containing 100 de-noised echograms generated from a device deployed on the surface of the water looking downward, in the Discovery Passage off Vancouver Island, British Columbia in 2015. Seventy echograms were used for training and 30 were reserved for testing. And while most indicated the presence of herring (as evidenced by a “strong intensity” core, a vertical elongated shape, and other features), all were pre-processed to remove undesirable signals such as small fish.
Above: Above: Echogram data processed by the researchers’ AI model. The black regions were classified as background, while red regions were classified as schools of herring. All detections were correct in the first two samples (left and center), while one was incorrectly classified in the right-most echogram.
The researchers report that the best-performing of three AI models bested the baseline method, albeit while incorrectly classifying some samples. Nevertheless, the team says that their system — which is scalable to other specifies like zooplankton and salmon — achieves performance guaranteeing “meaningful” detection.
They leave to future work simplifying the architecture to make it easier to expand to new detectable classes.
“We show that, even using a smaller annotated dataset, deep learning-based solutions can outperform a classical machine learning method. Our promising results indicate that with the annotation of more data, deep learning can be efficiently applied in underwater acoustic analysis,” wrote the coauthors. “The ability to measure the abundance of such subjects over extended periods of time constitutes a strong tool for the study of the effects of water temperature shifts caused by climate change-related phenomena.”
They aren’t the first to apply AI to ecology. Microsoft recently highlighted a Santa Cruz-based startup — Conservation Metrics — that’s leveraging machine learning to track African savanna elephants. Separately, a team of researchers developed a machine learning algorithm trained on Snapshot Serengeti that can identify, describe, and count wildlife with 96.6% accuracy. Intel’s TrailGuard AI system prevents poaching by detecting motion with an embedded camera using an offline, on-device AI algorithm. And scientists at Queensland University used Google’s TensorFlow machine learning framework to train an algorithm that can automatically detect sea cows in ocean images.