Environmental Prior Guided Signal Processing and Learning for Oceanographic Data Sensing and Analysis

We propose an approach which utilizes the underlying physical processes of the phenomenol-ogy to improve results and inform the ope'rators of the machine's decisions. This is particularly useful when understanding the delity of results in an environmental context. For example, imagine side-scan sonar collecting imagery in a shallow water environment with low sea-state.Such an environment likely results in imagery with multi-path contamination { poor perfor-mance of the remote sensing task. Being able to provide such context of the results to the operator serves to build trust and increase utility of future sensor deployments. The research interest of this proposal is to develop environmental knowledge guided regularization algorithms to mine the data in an intelligent manner. Our proposed deep networks will bene t from environmental prior information of many di erentavors. Examples include sensor positioning' within the water column and/or earth, environmental degradation models and seastate parameters from weather buoys etc.