FIELD TRAINABLE MISSION ADAPTABLE UNMANNED AIR VEHICLES

AIRCRAFT RECOGNITION PROCESSORS BASED ON NEURAL NETWORKS WHICH CAN BE LARGELY TRAINED 'ON THE GROUND' WITH 'TWEAKING'AT ALTITUDE ARE ENVISIONED. TWO RADAR SYSTEM STRATEGIES AREDESCRIBED BOTH UTILIZING A MULTI-WAVELENGTH ..., RADAR TO OBTAIN PART OF THE ASPECK ANGLE, 0, DEPENDENCE OF THE AIRCRAFT CROSS-SECTIONS, .... THE FIRST STRATEGY IS A NETWORK FOR RECOGNIZING THE 0 DEPENDENCE OF .... IT WILL EVALUATE RESTRICTED COULOMB ENERGY (RCE) AND BACKPROPAGATION (BP) AIRCRAFT RECOGNITION NETWORKS. TRAINING IMPOSES CALCULATED AND MEASURED ... PATTERNS ON NETWORK INPUTS WHILE MODIFYING LINK VALUES AND ENFORCING IDENTIFICATION OF THE AIRCRAFT AT THE OUTPUT NODES. RECOGNITION, WITH FIXED LINK WEIGHTS, IMPOSES FRAGMENTARY, NOISY RADAR ... PATTERNS ON THE NETWORK INPUT NODES WHILE OBSERVING OUTPUT NODE ACTIVATIONS. TRIAL NETWORKS WILL BE BUILT AND EVALUATED. THE SECOND STRATEGY IS AN INVERSE SYNTHETIC APERATURE RADAR (ISAR) USING NETWORKS TO ADAPT TO THE AIRCRAFT PATH AND COHERENTLY PROCESS THE TARGET RETURN TO ACHIEVE THE RESOLUTION OF A SYNTHETIC APERTURE. THE RESULTING IMAGE WOULD BE PROCESSED IN RECOGNITION NETWORKS AS IN THE FIRST STRATEGY. THE THEORETICAL REQUIREMENTS OF ISAR NETWORKS WILL BE STUDIED.