Hierarchical classification method and device

Abstract

The invention discloses a hierarchical classification method and a device. The method comprises the following steps: receiving input images via a neural network and identifying the images to output normalized scores; searching a maximum score in the normalized scores and determining the corresponding class; judging whether the maximum score is greater than a threshold or not; outputting the maximum score and the corresponding class as identification results when the maximum score is greater than the threshold; converting score distribution into rough class score distribution when the maximum score is not greater than the threshold, then circularly performing the steps of searching the maximum score in the normalized scores, determining the corresponding class and judging whether the maximum score is greater than a threshold or not, stopping circulating when determining that the maximum score is greater than the threshold, and outputting the maximum score and the corresponding class asthe identification results. Through the above manner, the hierarchical classification from fine to rough is achieved, so that the problem of wrong identification or missing identification caused whenall the fine classes of scores are low in the existing image identification task can be solved.

Description

technical field [0001] The invention relates to the technical field of neural network learning, in particular to a hierarchical classification method and device. Background technique [0002] Neural networks are widely used in the field of image recognition. In general, a trained neural network can only classify certain types of things, and requires that these categories are not included independently, such as apples and pears, and the scores of all categories need to be normalized ( All values ​​are between 0 and 1 and sum to 1). We can call these categories as fine categories, above which there are coarse categories, for example, apples and pears belong to pome fruit category, and then there are fruits category at the next coarse level, and so on. If there is a normalized coarse class output score, because the coarse class covers a larger range of categories, theoretically the coarse class score of the target should be higher than the included fine class score. [0003]...

AI Technical Summary

Problems solved by technology

[0005] The main technical problem to be solved by the present invention is to provide a hierarchical classification method and device to solve the technical problem of error and omission recognition caused when the scores of all subcategories are low in the image recognition task in the prior art

Images