Distributed Convolutional Neural Networks for IQ-based Multi-sensor Classification & Data Fusion
Abstract
ABSTRACT: Deep learning has become ubiquitous in its application towards computers learning difficult andcomplex problems. Their ability to learn their own features from raw data with very little or no a priori knowledge has become of increasing interest and Virginia Tech has been on the leading-edge of applying this to classification of raw IQ data. Our approach aims to perform sensing collaboratively in order to improve the classification and learning of each sensor pushing these techniques into SWAP-constrained electr"onic warfare platforms. In this way, the sensors become more independent over time thus reducing the overhead and reliance on the ne""twork with only short-term increases due to environment perturbations. Historically, most techniques towards aggregating sensor" classification results utilize a centralized classifier or consensus of the local results. The purpose of this proposed work is to develop a sensor network that reaches convergence and learns the RF environment using the learned feature vectors within the convolutional neural network. The diversity of the sensors working collaboratively from encoded feature sets can improve overall classific"ation rather than just trying to determine at the end what they should classify globally on locally, independent, and possibly flawe"ddata. It is expected that over time each sensor will reduce its dependence on collaboration and thecommunication of encoded feature vectors until the environmental changes induce a new learning cycle.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 01, 2017
- Source ID
- N000141712835
Entities
People
- Michael Fowler
Organizations
- Office of Naval Research
- United States Navy
- Virginia Tech