摘要 |
For automatic pattern recognition, a neural network has an input layer (IL) (two-dimensional field consisting of MxN elements, M = number of characteristic vectors, N = number of coefficients per characteristic vector) which is divided into overlapping pattern segments (LHxLV, with LH<M, LV<N), neighbouring segments overlapping partially at least in one direction. Each neural element of a downstream hidden layer (HL) is formed from a differently wieghted sum value of the amplitudes of a segment of the input layer (IL). The hidden layer (HL) is then fully networked with a one- dimensional output layer (OL), that element of the output layer which has the greatest sum value corresponding to the pattern to be recognised. <IMAGE> |