AI Utilizes Acoustic Data to Detect Herring Schools

2019-10-22

The impact of climate change on marine ecosystems can be understood by tracking the health of underwater species, but this is a time-consuming process that requires a lot of effort. Biologists need to use echosounders to study and use sonar to determine the depth of the water and the volume of objects, and they must manually interpret the two-dimensional echograms obtained. However, this is not the worst part; what is even more unacceptable is that manual interpretation is prone to errors. To fundamentally solve this problem, a group of researchers from the University of Victoria in Canada is developing a machine learning method for detecting specific biological targets in acoustic survey data. In order to improve the accuracy of environmental monitoring, they specifically chose herring schools for testing. In 2015, the team trained an artificial intelligence model in Discovery Passage off Vancouver Island, British Columbia, with a corpus of 100 denoised echograms—these images were generated by a device looking down from the water surface, with 70 echograms used for training and 30 for testing. The echogram experimental data from that year laid a solid foundation for the current development, saving researchers a lot of time and effort. The above figure shows three different AI training models, with the best-performing model surpassing traditional baseline methods. Although the success rate is not 100%, it fundamentally breaks through the previous constraints, which is undoubtedly a gratifying achievement. The team stated that their system can not only be applied to herring schools but can also be extended to other specified planktonic organisms, ensuring accurate and meaningful detection tasks while demonstrating superior performance. "We have demonstrated that even with a very small annotated dataset, deep learning-based solutions can outperform traditional machine learning methods. Our research results indicate that with more annotated data, deep learning can be effectively applied to underwater sound analysis. And for a long time, it has been a powerful tool for studying the impact of climate change-related phenomena on water temperature changes by measuring the quantity of such subjects," the researchers wrote in their paper. The University of Victoria in Canada is not the only university applying artificial intelligence to ecology. Recently, scientists from the University of Queensland trained an algorithm using Google's TensorFlow machine learning framework to automatically detect dugongs in marine images. In addition to universities, Conservation Metrics, a startup based in Santa Cruz, is using machine learning to track African savannah elephants. Furthermore, Intel's TrailGuard AI system recently integrated an offline, on-device AI algorithm that uses embedded cameras to detect movement, effectively deterring poachers and protecting wildlife.

RECENT AI NEWS

RECENT AI TOOLS