In-situ analytics for tomographic imaging in sensor network

In both industry and academia, the seismic exploration does not yet have the capability of illuminating the physical dynamics with high resolution and in real-time. The major bottleneck in real-time monitoring today is to transfer large volume of raw data for post processing. Although computation capacity and sampling rate of sensors have increased exponentially, we still have challenges in terms of communication and battery life. To monitor physical dynamics in real-time and to avoid costly data transfer, we need to perform in-situ computation and analytics. In this paper, we present a decentralized least-squares solver that can perform in-network computation and generate tomography image in real time. The proposed solution is evaluated using both synthetic and real data set. Preliminary evaluation shows that the decentralized method can recreate the image close to the one obtained from the centralized computation. We envision the system can be applied to a wide range of seismic exploration topics such as hydrothermal, oil exploration, mining safety, mining resource monitoring. The scientific and social impact is broad and significant.