| 摘要 |
PROBLEM TO BE SOLVED: To provide a multi-class discrimination device capable of avoiding redundant learning in classification of multi-class images including images closely resemble with each other or data having other features.SOLUTION: The multi-class discrimination device is constituted of double-layered discrimination device including: a first layer discrimination device 208 that performs classification of multi-class images including images closely resemble with each other or data having other features in a group unit; and a second layer discrimination device 210 that performs classification within each of groups. When determining groups, a first learning device 201 previously generates a category discrimination unit 202 for performing discrimination in a state of non-grouping using machine learning. A discrimination error totaling device 204 performs a discrimination test using the category discrimination unit 202 to total the number of discrimination errors among several categories. A grouping processing unit 206 performs grouping of categories which are easy to cause errors in machine learning. Closely resemble items are automatically sorted by previously totaling discrimination errors and thus double-layered discrimination including an intergroup discrimination and an in-group discrimination is executed. Accordingly, multi-class discrimination device reduces redundant learning. |
