Depth image clustering method, system and device, medium and terminal

Abstract

The invention belongs to the technical field of unsupervised image recognition, and discloses a depth image clustering method, system and device, a medium and a terminal.The depth image clustering method comprises the steps that a set of pictures is input, and multiple sets of codes of each picture are obtained through an encoder in an SIMO self-encoder; the multiple groups of codes of each picture and the codes obtained after information fusion is carried out on the codes are input into the same decoder to obtain reconstructed pictures; the reconstruction loss of the reconstructed picture and the picture obtained after data enhancement of the input picture is calculated, and pre-training of the network is carried out; after the pre-training is finished, the fusion codes are clustered to obtain comparison clustering loss, the comparison clustering loss is added into the training process for joint training, and parameter optimization is achieved; and after the joint training is finished, feature extraction on the whole data set is performed to obtain a fusion code, and the fusion code is input into a traditional clustering algorithm to obtain a clustering result. According to the method, semantic information of multiple angles of a single image can be extracted at the same time, and the richness of the information is improved.

Description

technical field [0001] The invention belongs to the technical field of unsupervised image recognition, and in particular relates to a deep image clustering method, system, equipment, medium and terminal. Background technique [0002] At present, with the rapid arrival of the big data era, hundreds of millions of image data are generated on the Internet every day. If the information of massive image data can be fully utilized and mined, it will be able to generate extremely high commercial value. For example, image recognition is a specific application scenario for mining the value of images, and most traditional data mining algorithms cannot directly process image data, so people often manually calculate various features of images and then use these algorithms for further identification. [0003] In recent years, with the development of computer vision technology, image recognition technology based on deep learning has developed extremely mature, especially in the case of im...

AI Technical Summary

Problems solved by technology

[0006] (1) In the existing technologies based on unsupervised image recognition, the structure at this stage is just to compress the image blindly into a set of codes. From a physical point of view, this set of codes often can only represent a certain angle of the image. information, which means that a large amount of information loss is caused in the process of feature extraction

[0007] (2) The significant assumption of the clustering loss based on Euclidean distance at this stage is that the output of the code needs to be separable in Euclidean space, but the neural network is often nonlinear, so this assumption is difficult to satisfy, and it is difficult to ensure that these Can the effect of clustering be significantly improved after loss?

For (2), the difficulty lies in how to use a clustering loss to ensure the improvement of the clustering effect under the assumption that the encoding does not satisfy the Euclidean spatial distribution

Images