Image retrieval method based on deep convolutional neural network

A deep convolution and image retrieval technology, applied in the field of deep convolutional neural networks, can solve the problem of inability to solve the problem of too much storage space for computational data, and inability to learn deep-level features of images, etc., to achieve the effect of improving accuracy

Inactive Publication Date: 2017-08-25
桂林明辉信息科技有限公司
View PDF3 Cites 49 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is: in the prior art, in the method of combining the deep convolutional neural network and the hash technology, there is no way to learn the deep-level features of the image, and there is no way to solve the problem that the calculation data volume is too large and the storage space is too large. The problem

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
  • Image retrieval method based on deep convolutional neural network
  • Image retrieval method based on deep convolutional neural network
  • Image retrieval method based on deep convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0048] Such as figure 1Shown, a kind of image retrieval method based on deep convolutional neural network of the embodiment of the present invention, this method comprises the following steps:

[0049] The first step is to collect image data, perform normalized preprocessing on the collected image data, and then divide the preprocessed image data into training set image data and test set image data, and divide the training set image data and test set image data The image data is stored in the image database; among them, the collected image data (such as MINST and CIFAR-10 data sets, the MINST image data set is composed of 7000 grayscale pictures, and the content of the pictures is handwritten 0 to 10; CIFAR-10 image The data set is composed of 10 categories of 32*32 color images. Download three experimental data sets from the Internet, delete the noise pictures first, perform normalization processing, and perform normalization preprocessing on them, and normalize The preproce...

Embodiment 2

[0078] Such as figure 2 with image 3As shown, the embodiment of the present invention is a deep convolutional neural network model, including: sequentially connected input layer, convolution layer, sub-sampling layer, magnetization layer, fully connected layer, the model includes: hash layer, loss layer and the retrieval layer; the hash layer, the loss layer and the retrieval layer are sequentially connected behind the fully connected layer; the hash layer includes: sequentially connected slice layer, fully connected layer, activation layer, merging layer and thresholding layer The retrieval layer, which includes: a coarse level retrieval layer and a fine level retrieval layer connected in sequence; the loss layer, which uses functions including: Softmax classifier loss and quantization error loss; the retrieval layer, It includes: a coarse-level retrieval layer and a fine-level retrieval layer connected in sequence; the coarse-level retrieval layer is used for coarse-level...

Embodiment 3

[0080] ST1: Collect short image data (such as ImageNet, MINSTHE and CIFAR-10, etc.) from the Internet, preprocess the collected data sets, including image denoising, etc., and divide the data sets to be used for retrieval into The training set and the test set store the processed image data in the image database.

[0081] ST2: Construct a deep neural network framework model. The model of the present invention is based on the convolutional neural network model. The seventh layer of the original neural network model and the replacement layer hash layer are used as the new seventh layer, and after the seventh layer Add a loss layer to calculate the quantization error, use the backpropagation algorithm to fine-tune the network for different data sets, and optimize the target.

[0082] The first is the convolutional neural subnetwork, which includes convolutional layers, downsampling layers, and fully connected layers. Use the powerful learning ability of convolutional neural netw...

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 relates to an image retrieval method based on a deep convolutional neural network. The method comprises the following steps that: S1: collecting image data; S2: constructing a deep convolutional neural network model; S3: after the model is trained and subjected to parameter regulation, learning binary Hash encoding and calculating losses in the model; and S4: after the losses are calculated, carrying out image retrieval on test set image data. Through the method, the deep characteristics of an image are learnt, image retrieval accuracy is improved, and the problems that the deep characteristics of the image can not be learnt and an overlarge data size storage space can not be solved are solved.

Description

technical field [0001] The invention relates to the field of deep convolutional neural networks, in particular to an image retrieval method based on deep convolutional neural networks. Background technique [0002] With the development of the Internet and the wide adoption of image acquisition equipment, people can easily obtain images, but the number of images is also increasing rapidly, which puts a huge pressure on the computing power and storage efficiency of the system, especially in such a huge It is especially difficult to retrieve data quickly and efficiently on large datasets. How to quickly and efficiently retrieve large-scale image resources to meet user needs needs to be solved urgently. Image retrieval technology has gradually developed from early text-based image retrieval (Text-Based Image Retrieval, TBIR) to content-based image retrieval (Content-Based Image Retrieval, CBIR). The visual low-level features include gradient-based image local feature descripto...

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 Applications(China)
IPC IPC(8): G06F17/30G06N3/04G06N3/08
CPCG06F16/5838G06N3/084G06N3/045
Inventor 黄文明杜梦豪魏鹏
Owner 桂林明辉信息科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap