摘要 |
PURPOSE:To collect good training data with less operation by providing a non- linear conversion module to perform the non-conversion, and observing (n) dimensional feature amount data distribution by one-time non-linear conversion. CONSTITUTION:A non-linear conversion module 2 performs non-linear conversion of training data inside a data base 1. These converted data displayed on a graph and list of a display module 3 and a single non-linear conversion enables the observation of (n) dimensional feature amount data distribution. Next, a correction module 4 corrects the base 1, by which the specified part of training data is removed or returned from a non-linear conversion display as an object. A control module 6 controls activation and completion of each module with respect to mankind's request, which enables smooth repetitive processing such as non-linear conversion, correction, etc. In such a way good training data is collected by less operation. |