APPLICATION OF A LEARNING RECOGNITION SYSTEM FOR THE SEPARATION OF AN UNKNOWN SIGNAL FROM NOISE,
FOREIGN TECHNOLOGY DIV WRIGHT-PATTERSON AFB OHIO
Pagination or Media Count:
An experiment in the detection of a signal with its form not given in advance in the presence of great noise with the aid of a learning recognition system L-optimal recognition system is described. Pictures composed of a random accumulation of points appear on a 10 x 10 cm screen. The number of points varies from 20 to 40. Each picture belongs to one of two classes. The picture belonging to the first class contains a signal in the form of a rhomb of eight points. The dimensions of the rhomb in all pictures of the first class are the same, only their locations and orientations change. The pictures of the second class are not rhombs and imitate the noise without a signal. The figures of the first class imitate the signals in the presence of such a great noise that when the form of the signal is submitted to a man he makes a 5 percent error in recognizing it when the time of recognition is limited to 2-3 sec. During the experiment the form and the character of the signal is not submitted to the recognition system or to a man, and only the pictures of both classes instruction sequence, indicating to which class they belong are shown.