Maritime Sensing Research of the Environment, Features, Objects and Activities (EFOA)
Abstract
Maritime Sensing Research of the Environment, Features, Objects and Activities (EFOA)""With the proliferation of electro-optical (EO"), hyper-spectral and microwave (synthetic aperture radar) sensors being launched on small satellite constellations and legacy systems on multiple orbit planes, the strength of remote sensing to observe persistently energetic flow features and geophysical processes at the sub-mesoscale is gaining much momentum. In oceans, marginal ice zones and littoral environments, salient features such as currents, fronts, plumes, filaments, vortices, wakes, wave-current interactions and internal waves, as well as ocean-ice interactions and ice floes may exist at different spatial scales. Satellite constellations can now provide several times daily spatial snapshots of oceanographic phenomena. Using a constellation of different sensors and their daily imaging times will lead to near persistent coverage of nearly all locations on earth. A combination of high resolution images (~1 meter) will produce large spatial mosaics and from repeat orbit observations coherent change detection to develop an improved understanding of how signatures of oceanographic phenomena are related to kinematic and dynamic processes.Exploring the use of multi-polarizations, frequencies and sensor types will enhance the detection and observations of many processes occurring in the highly dynamic oceans and littoral zones. With the prospect of hundreds of satellites in space observing the ocean daily and soon continuously, the data sets derived from these images will be large (~10~s-100~s TB daily) creating unprecedented spatial and temporal resolutions that may challenge numerical modeling techniques. An effective approach to analyzing large volume of satellite image data is with MLTs which are trained to capture and extract invariances and consistencies from large, four-dimensional geophysical data sets. The MLTs can be applied to satellite data and data output from numerical simulations to test if the physics and numerics are correct in the models. Suitable Machine Learning Techniques (MLTs) with appropriate architecture and learning processes are needed to enable the recognition of features, objects and activities (FOAs) in different environments that will enable improved detection and classifications of FOAs and environmental predictions in numerical models. The number of sensor modalities (satellite to drones) are rapidly increasing and classical, fundamental algorithms need to be transported to Artificial Intelligence (AI) approaches for efficient and effective analytics of large volume datasets from a variety of sensor platforms. Multi-sensor satellite imaging alone will provide large datasets of oceanographic features and processes in different environments such a deep blue ocean, littoral environments and marginal ice zone for analyzing and feature and object extraction by MLTs. Numerical simulations will generate fine-resolution, physics-based data complementary to satellite remote sensing measurements. The comprehensive data obtained from the various sources will then be used to train the machine learning (ML) method for signature extraction of internal waves, currents, fronts, and ice floes and other phenomena. Improved modeling approach"es will allow to explore advanced direct analytics, data assimilation techniques and data fusion approaches.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 20, 2019
- Source ID
- N000141912514
Entities
People
- Hans Graber
Organizations
- Office of Naval Research
- United States Navy
- University of Miami