Real-Time Hybrid Neural Net Object Recognition System (HNNORS)

Recognition of minw-like targets using an autonomous underwater vehicle (AUV) is a difficult task due to algorithm complexity and data volume. Presently, much of the video data gathered in the field has to be processed Off-line. Computer system that can perform billions of floating point operations per second (GFLOPS) are required for on-line processing of 512x512 images. ORINCON proposes to design a Hybrid Neural Net Object Recognition System (HNNORS) by integrating a high-performance parallel processor architecture with advanced neural net classifiers and a suite of image processing algorithms. Four tasks will be completed in the Phase I effort: (1) Conduct an extensive investigation into the available high-performance parallel processor architectures based on commercial off-the-shelf (COTS) hardware and then propose a final candidate; (2) Perform rapid prototyping of HNNORS using the Programming and Processing Environment (RIPPEN) algorithm proven successful for underwater scene hardware modules into the parallel a contributions to enhance HNNORS, real-time performance prototype HNNORS on an in-house parallel processor video imagery. 93) Conduct an investigation into the incorporation of available dedicated image processing hardware modules into the parallel architecture and evaluate their potential contributions to enhance HNNOR's real-time performance; and and (4) Demonstrate a architecture using underwater video imagery.