JMesh -- A Scalable Web-Based Platform for Visualization and Mining of Passive Acoustic Data

摘要

Visual presentation of the outputs from marine mammal detectors is key for efficient mining of large acoustic datasets. Operators must be able to easily navigate through large time series of detections, examine spectrograms, listen to detected sounds, and validate and compare detections for different species over time and space. The JMesh web platform has been designed with these constraints in mind. The interface is organized around three interconnected visualization panels: 1) a geographic interface displays maps showing the total number of detections for each species at all monitoring locations within an adjustable time period, 2) a detection time series plot displays temporal variations of detections for several species at a selected monitoring location, and 3) a multimedia panel allows the user to visualize spectrograms, listen to sounds and validate detections. All three panels are interactive and allow the user to navigate intuitively between them. The platform uses load balancing, microservices orchestration, shared non-relational databases, and virtualization technologies to make the infrastructure fully scalable and expandable from a single server to a resource farm composed of hundreds of hosts. JMesh can display detections from archived data or from real-time acquisition systems. Compatibility with mobile devices through the Bootstrap framework simplifies access to the data while in the field.