A Dimensionality Reduction Method Based on Convolutional Neural Networks and Covariance Tensor Matrix

Abstract

The invention discloses a dimensionality reduction method based on a convolutional neural network and a covariance tensor matrix. The image is input to the convolutional neural network to extract the shape features of the edge image; in order to enrich the image detail features, the convolutional neural network is used to extract the original image features; the traditional feature extraction method is combined with the convolutional neural network feature extraction to obtain multiple a visual characteristic. The present invention pays attention to the shape features of the image target, and uses the convolutional neural network to extract the image features. Compared with the traditional feature extraction method, it can express the image more abundantly and intuitively. The correlation between them can make it more robust and practical to represent images as a whole.

Description

technical field [0001] The invention relates to the field of pattern recognition and machine learning, more specifically to a dimensionality reduction method based on a convolutional neural network and a covariance tensor matrix, and belongs to the technical field of data dimensionality reduction. Background technique [0002] In the era of big data, people's ability to collect and obtain data is getting stronger and stronger. Big data is infiltrating into various fields in today's world in various forms, such as biological gene functional group information, text classification and picture multimedia, and these data have shown a large amount of data, high dimensionality, heterogeneity, dispersion and structure Complexity and other characteristics, the massiveness of data will cause problems such as high storage overhead and slow retrieval speed; the high-dimensionality of data will cause the problem of dimensionality disaster, and the expanded dimension will lead to a rapid ...

AI Technical Summary

Problems solved by technology

[0004] Therefore, various dimensionality reduction algorithms have received extensive attention from researchers, and researchers are eager to find suitable dimensionality reduction methods to solve the problems of large storage capacity and high computational complexity. However, there are still some problems in the existing dimensionality reduction methods: (1) when dealing with When using an image, the shape is an important clue to confirm the image target, and most dimensionality reduction methods tend to ignore the shape features of the target in the image when processing image data; (2) only focus on a certain type of feature of the image, but ignore Other features make it impossible to express images in a rich and comprehensive way, and it is impossible to represent image data with multiple visual features as a whole

Images