THE MANTA ROVER: AN AUTONOMOUS AQUATIC
MONITORING AND SHADING SYSTEM FOR CORAL
Duration: Summer 2018
Location: New York University, Tandon School of Engineering
Research Director: Dana Karwas
Research Assistant: Diego Eduardo Kleiman (NYUAD) Computer Science
Research Assistant: Eva Zheng (NYU Tandon) Computer Science
Abstract: The Manta Rover system introduces new potential geo-engineering design experiments for local solutions to climate change. It brings up questions concerning artificial intelligence and experimental conservation technologies, and highlights using species-centered design as a way to consider small-scale solutions to problems created by climate change. By using targeted shading to mitigate effects of thermal stress to prolong the survival of coral communities, the Manta Rover introduces temporal mitigation measures until solutions to address global climate change become more effective (1). The Manta Rover filters the light that reaches the reef and uses image analysis with visible and hyperspectral bands to identify threatened coral ecosystems. The Manta Rover is programmed to automatically shade locations that are identified as threatened by the deep learning system that views the reef in real time using cameras attached to the floating shade. The system also identifies when the reef is healthy and the Manta Rover can move (through a GPS and propulsion system) and anchor to another location. The shading system is composed of an “optical fabric” that floats on top of the water and creates shade but is composed of dichroic filters, only letting the visible spectrum pass through (400 - 700 nm).
The project’s goal is to provide a proof of concept of the Manta Rover by demonstrating the feasibility of implementing two essential components of the system: optical filters and machine learning models for coral recognition. The optical filters were tested in a small-scale experiment, employing artificial lighting, samples of the coral Pocillopora damicornis, and dichroic filters to assess the benefits of limiting the wavelengths on the coral. A deep neural network capable of automatically classifying coral images was developed using the BENTHOZ-2015 public dataset and the VGG16 architecture (2, 10).
A case study aquarium experiment was set up using different light spectrums (full spectrum and visible spectrum) upon healthy specimens of the coral Pocillopora damicornis. We analyzed the effects of light stress by evaluating color changes using pixel-based image analysis of a series of photographs taken every ten seconds for the duration of the experiment. This experiment was designed to identify effects of manipulating the visible spectrum upon the coral as a way to partially limit the light frequencies, and identify the ideal limited spectrum that would compose the filtering system embedded in the larger Manta Rover floatation system.
Thanks to the NYU Summer Research Program for Support