A data-oriented deep clustering method for sole pattern images

Abstract

The present invention provides a data-oriented deep clustering method for sole pattern images. Based on the self-expression layer of the unsupervised encoding network for sole pattern images, the sole pattern features learned during the self-encoding process can be used as shoe samples or suspects with category information. Sole pattern feature training, through the pre-training of shoe sole pattern image datasets with different attributes, you can use the supervised network to define the sole pattern feature subspace to restrict the clustering process of unlabeled on-site shoe sole patterns or suspect shoe sole patterns , so that the clustering process of the sole pattern is evidence-based. In addition, the present invention provides a pre-training strategy for pattern image feature subspace training based on the order of training models for different pattern data sets, which more effectively reflects the process of label-free clustering training of shoe sole patterns. Constraint effect, improve clustering accuracy. This clustering method can meet the clustering and classification processing of unbalanced data sets.

Description

technical field [0001] The present invention relates to a clustering method, in particular, to a data-oriented deep clustering method of sole pattern images. Background technique [0002] At present, in the clustering algorithm system of sole pattern images, there are mainly the following two types: [0003] 1. The fuzzy clustering algorithm based on multi-label, the main idea is: use the local or overall features of the sole pattern image to calculate the similarity of the sole pattern image, perform single-label clustering according to the similarity between the pattern images, and then according to each pattern image. The similarity between the class and the sole pattern is correlated between the classes, so as to achieve the purpose of multi-label clustering. For reference, see the patent "A Multi-label Clustering Method of Sole Pattern Images", patent number CN201710446061.4. [0004] 2. Non-fuzzy clustering algorithm, the main idea is to extract the local or overall ...

AI Technical Summary

Problems solved by technology

Based on the principle of transfer learning, the present invention uses various shoe sole pattern data itself as a bridge for deep clustering, and based on an unsupervised coding network or generation network, so that the coding features of shoe patterns and suspect shoe sole patterns can be applied to classification while taking into account Deep clustering, based on the training of sole pattern data sets containing classification marks, can solve the problem of uneven distribution of supervised data sets in the learning process, such as shoe sole patterns, unsupervised pattern learning through potential classification training based on other categories of pattern images, The data set can be used as a guide to solve the clustering problem of on-site patterns and improve the clustering accuracy

Images