Sensory Stream Data Mining on Chip

Mining physical properties from real-time sensor stream data is important to the atmospheric studies, ecology and oceanography. An FPGA-based reconfigurable sensory stream data mining processor is presented in this paper. The processor is based on Generalized Non-Linear Regression algorithm and trained with radiative transfer simulations and observations for autonomous detection of satellite measurement signatures and retrievals of atmospheric physical properties. The results show that the embedded computing approach is faster than traditional computing methods in orders of magnitude. The data mining processor is able to automatically adapt to multi-platform sensors, eliminate redundant algorithm development and provide a vehicle for autonomous onboard image analysis and physics-based data compressions.