This feasibility project will define, design and prove relevant concepts on the use of Acoustic Charge Transport (ACT) technology for implementing neural-network based image target recognition. ACT is the only technology available today which promises a level of performance equivalent to 46 billion multiply-adds per second on a single IC chip. The proposed system may be capable of very high throughput, up to 200 million pixels per second, typical of synthetic aperture radar and laser based image detection systems. We will investigate various neural network modeling paradigms, configure an ACT device to perform the neural network modeling of scanned imagery and develop whatever necessary additional software and hardware for this task. If successful, we will have created the world's fastest non-optical neural net pattern recognizer of raster scanned images. Anticipated benefits/potential commercial applications - the primary benefit we expect to attain is an assessment of how well ACT technology can perform neural-network based image processing at extremely high speed, suitable for synthetic aperture radar and laser based target recognition.
