Hyperspectral and Persistent Sensor Signal Processing Platform and Algorithms

ABSTRACT:There is a growing need for timely processing, exploitation, and distribution of sensor data collected by unmanned aerial vehicles to generate actionable intelligence. Challenges exist in producing hardware that is rugged enough to withstand the physical demands of airborne platforms, yet able to deliver significant computing performance within the size, weight and power constraints of smaller airframes. CG proposes to create an Embedded Avionics High Performance Computing (EA-HPC) system that will meet this need. This will include hardware for sensor capture, raw data storage, sensor data processing, and interfacing with the aircraft communication system. The system will also contain software to enhance the raw data, detect potential targets, fuse sensor inputs, and track targets using midwave infrared and hyperspectral imaging sensors. The innovation present in this design will be to provide choices between three state-of-the-art processing elements: (1) Conventional general purpose Central Processing Units (CPUs), (2) Modern CUDA-compatible Graphics Processing Units (GPUs), (2) Power efficient and flexible multicore accelerators. The solution will scale easily from one to multiple sensors with minimal effort. This will allow a wider variety of missions to be executed without major refitting.BENEFIT:This approach will produce a platform with high processing efficiency in GFLOPS/Watt, and the scalability to adapt to different sensor configurations in addition to developing software to exploit the sensor data. The systems small footprint will contain a powerful assortment of computing capabilities that can accelerate a variety of sensor processing algorithms. The onboard processing of vehicle tracks will make it possible to send the tracking results to the ground station instead of all the imagery. This reduces the communication bandwidth required while increasing the usability of the transmitted data.