DANIEL H WAGNER ASSOCIATES INC — Department of Defense SBIR Phase I: N231-038

Description

The critical role Unmanned Underwater Vehicles (UUV) play in Naval operations is constrained by their ability to send large volumes of sensor data to warfighters. Thus, compression technology which retains all features of the data needed for human analysis, called perceptually lossless compression, is required. Additionally, our Autoencoder Compression Enhancement for UUV Sensor Data (ACE4USD) system must operate with error prone communication links. To do this, ACE4USD will utilize Deep Neural Network (DNN) based autoencoders to extract key features of UUV sensor data. ACE4DSU will choose features which are substantially smaller than the observed sensor data, but nevertheless permit perceptually lossless reconstruction of the sensor data. ACE4USD will exploit unsupervised learning techniques for DNNs which make these features amenable to further compression by conventional methods. ACE4USD will also use DNN denoising autoencoders to deliver perceptually lossless sensor data to warfighters in the presence of error prone communication links.