No abstract

No abstract

A detailed understanding of ecological soundscapes provides a window into ecosystem state that might otherwise go undetected. Rather than consider acoustic emissions of a particular species, a top down ecosystem acoustics point of view gives potentially valuable insights to reef ecosystem function. Simultaneous passive acoustic and video data were collected from coral reefs off the Kona Coast of Hawaii in January 2019 during a soundscape experiment. Frequent interactions between fish species as well as collective behaviors are identified and tracked by applying computer vision to video data. Comparison relating video data to acoustic data allows assessment of the same behaviors as simultaneously reflected in directional passive acoustic data. Collective consideration of acoustics and video offers insight into acoustic signatures of particular soniferous communities and behaviors amongst and between reef species. Machine learning applications match patterns between fish trajectories in videos and the sound field as received by a directional antenna. The relative level of biological activity and presence or absence of particular behaviors provides key data in the effort to understand coral reef ecosystem state.

Generative adversarial networks (GANs) have shown tremendous promise in learning to generate data and effective at aiding semi-supervised classification. However, to this point, semi-supervised GAN methods make the assumption that the unlabeled data set contains only samples of the joint distribution of the classes of interest, referred to as inliers. Consequently, when presented with a sample from other distributions, referred to as outliers, GANs perform poorly at determining that it is not qualified to make a decision on the sample. The problem of discriminating outliers from inliers while maintaining classification accuracy is referred to here as the DOIC problem. In this work, we describe an architecture that combines self-organizing maps (SOMs) with SS-GANS with the goal of mitigating the DOIC problem and experimental results indicating that the architecture achieves the goal. Multiple experiments were conducted on hyperspectral image data sets. The SS-GANS performed slightly better than supervised GANS on classification problems with and without the SOM. Incorporating the SOMs into the SS-GANs and the supervised GANS led to substantially mitigation of the DOIC problem when compared to SS-GANS and GANs without the SOMs. Furthermore, the SS-GANS performed much better than GANS on the DOIC problem, even without the SOMs.