A Deep Enhanced Image Clustering Method

Abstract

The present invention proposes a deep enhanced image clustering method, belonging to the technical field of image clustering and data mining, 1) pre-training codec network, initializing potential feature space; 2) adopting traditional K-means method to cluster in potential feature space Class centroids are initialized, and Bernoulli-logistic units are assigned to each centroid; 3) Logistic regression parameters and Bernoulli distribution between the point and the unit are calculated; 4) Temporary rewards are dynamically allocated using the reward regression strategy, and joint auxiliary goals Distributed calculation of the motion trajectory of each centroid; 5) Calculate the weight, iteratively optimize the clustering unit until the convergence condition is met, and complete the depth-enhanced image clustering process. At the same time, the present invention is based on the idea of ​​reinforcement learning and uses the reward regression strategy to jointly use latent features to represent and adjust the cluster centroid, so as to fully apply all the cluster information, especially the cluster information in the adjacent area, to the process of cluster analysis. In the interaction between environment and behavior Effectively improve the problem of clustering fuzzy, and effectively improve the clustering performance.

Description

technical field [0001] The invention belongs to the technical field of image clustering and reinforcement learning, and relates to a deep image clustering method based on reinforcement learning. Background technique [0002] With the rapid development of Internet of Things technology and network information technology, the popularity of smart phones, tablet computers and other electronic products is becoming wider and wider, and more and more data can be collected, and the data structure is becoming more and more complex, especially the amount of unstructured image data. Even more explosive growth. Image data contains rich semantic information for research in various fields, but due to the complex structure and high dimensionality of the data, it is difficult to accurately obtain the rich semantic information in the data. Therefore, it is urgent to study a new method to deeply mine the rich information in massive image data. [0003] Clustering often conducts data analysis...

AI Technical Summary

Problems solved by technology

However, the existing deep clustering algorithms lack the consideration of the entire clustering environment, especially the influence of the surrounding area environment on the clustering effect. The clusters of some image input points are relatively fuzzy

Images