PURPOSE: To improve the rate of identification while approximating the automatic identifying method of a computer to biological one by inserting preprocessing such as wavelet transformation for arbitrary data such as image data or time sequential data and identifying those data while increasing the feature quantity of them.
CONSTITUTION: In the case of observing one object like (a,) for example, a human eyeball 101 simultaneously detects both of a rough feature and a fine feature with an X cell 102 and a Y cell 103 and identifies that object. The arbitrary data such as image data or time sequential data are not conventionally identified by a neural network as they are or stepwise while applying this principle but identified while providing the several kinds of observation by inserting preprocessing 105 such as wavelet transformation like (b). Further, when any parallel computer is employed, parallel processing is enabled as the feature of the neural network, and processing speed is biologically approximated.
MATSUBA IKUO
HITACHI ADVANCED SYST KK