Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A neural network feature learning method based on image self-encoding

A neural network and feature learning technology, applied in the field of image retrieval and deep learning, can solve problems such as the limitation of neural network expression ability, achieve the effect of improving semantic expression ability, solving insufficient structural information, and improving accuracy

Active Publication Date: 2020-04-28
BEIJING UNIV OF TECH
View PDF8 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the general lack of supervisory information in deep neural networks has limited the expressive ability of neural networks.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A neural network feature learning method based on image self-encoding
  • A neural network feature learning method based on image self-encoding
  • A neural network feature learning method based on image self-encoding

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0026] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings. figure 1 It is the overall flowchart of the method involved in the present invention, with figure 2 It is a general structure diagram of the algorithm involved in the present invention.

[0027] Step 1: Construct the dataset

[0028]The database in the implementation process of the method of the present invention comes from two public multi-label standard data sets PascalVOC 2012 Segmentationclass and Microsoft COCO. Among them, Pascal contains 1,465 training, 1,449 testing, and the total number of categories is 20 categories of color pictures; Microsoft COCO contains 82,783 training, 40,504 testing, and category summary is 80 categories of color pictures. The segmentation labels corresponding to the image training set are respectively represented on the original image, and the main objects of each graph will be marked with different colors w...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a neural network feature learning method based on image self-encoding, and belongs to the technical fields of feature learning and image retrieval. Firstly, the segmented training image set corresponding to the training image set is constructed through the segmented labels of the multi-label image dataset, and then the weights of the convolutional neural network and the self-encoder neural network are initialized, and the self-encoder neural network is trained using the stochastic gradient descent method, and each training image is extracted. The samples correspond to the latent variables of the segmented image and are normalized. Then, use this hidden variable as the training target corresponding to the original image in the training set, train the convolutional neural network, and extract the feature vector corresponding to each image in the test set image library, by calculating the query image and each image in the image library The Euclidean distance between the eigenvectors, and the distances are arranged in ascending order to obtain similar image retrieval results. The invention enables the features extracted by the trained neural network to achieve more excellent retrieval effect on the multi-label retrieval task.

Description

technical field [0001] The invention relates to the fields of deep learning and image retrieval, in particular to a feature expression method in image retrieval, which can obtain more accurate similar images on a multi-label data set. Background technique [0002] With the development of multimedia and network technology, images, as the most intuitive way of expressing people's living conditions, play an increasingly important role in people's lives. Most images contain rich semantic information, how to find the images that users need in real life is a difficult problem and challenge. Excellent feature expression can not only represent the category information of the image, but also capture the relevant semantic information of the image. A large amount of image information is collected and utilized. However, combining image processing with computer vision technology to extract effective semantic expressions in images is the top priority in the field of computer vision. How...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/66G06K9/46G06N3/08G06F16/583
CPCG06F16/5838G06N3/084G06V10/464G06V30/194
Inventor 段立娟恩擎苗军乔元华
Owner BEIJING UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products