Connectionist Model for Automatic Target Recognition Based in "Object Recognition by Parts"

Dr. Jakubowicz has nine years experience in neural network research and development for the purpose of object and target recognition. In this proposal a work plan is put together for a system to perform Automatic Target Recognition utilizing primitive subobjects as components of the complex objects. This will include devising an integrated model for recognizing complex objects with component identification at an intervening level. The major modules of the model are already constructed at Jireh Systems. A large part of the Phase I project will be aimed at integration of the modules. The model will be implemented on a workstation for a small problem set and demonstrated in Phase I. A final report will describe the resultant model and demonstration results. An important component of any such system is the intelligent sequential control of parallel recognition processes as demonstrated in higher vertebrate animals. Neural networks are proposed for the parallel recognition process while both neural networks and rule based systems are being considered for the sequential control processes.