High Quality/Low Dimension Data for Sensor Integration

Novel algorithms were developed for ecient multi-sensor fusion of correlated high-dimensional data to support target detection. In our compressive sensing framework, high-dimensional data is compressed using low dimensional random projection matrices. When the high-dimensional data exhibits low-dimensional structures, this scheme is capable of capturing all the significant informative with fewer random projections. Copula theory permits exploitation of correlation between sensors to significantly enhance system performance as compared to that of traditional systems that assume independence. Testing was performed on multisensor data sets including physics-based simulations and real-world seismic data. The performance of the algorithms was compared to benchmark algorithms. The results demonstrate our approach is superior to current methods for both individual sensors and multiple fused sensors. Furthermore, our fusion results demonstrate performance gains for multiple sensors compared to single sensors. The approach provides significant reductions in communication bandwidth requirements compared to systems that transmit raw sensor data to a central fusion center. In Phase II, we propose to enhance the current algorithms and adapt the approach to additional sensor types. We will work with our transition partners to derive requirements and test performance on data from systems of interest to the DoD and commercial markets.Novel algorithms were developed for ecient multi-sensor fusion of correlated high-dimensional data to support target detection. In our compressive sensing framework, high-dimensional data is compressed using low dimensional random projection matrices. When the high-dimensional data exhibits low-dimensional structures, this scheme is capable of capturing all the significant informative with fewer random projections. Copula theory permits exploitation of correlation between sensors to significantly enhance system performance as compared to that of traditional systems that assume independence. Testing was performed on multisensor data sets including physics-based simulations and real-world seismic data. The performance of the algorithms was compared to benchmark algorithms. The results demonstrate our approach is superior to current methods for both individual sensors and multiple fused sensors. Furthermore, our fusion results demonstrate performance gains for multiple sensors compared to single sensors. The approach provides significant reductions in communication bandwidth requirements compared to systems that transmit raw sensor data to a central fusion center. In Phase II, we propose to enhance the current algorithms and adapt the approach to additional sensor types. We will work with our transition partners to derive requirements and test performance on data from systems of interest to the DoD and commercial markets.