Unbalanced trademark retrieval method based on deep hash of aggregation similarity

A similarity and trademark technology, applied in the computer field, can solve problems such as imbalance, data imbalance, and low accuracy of retrieval results, and achieve the effect of solving retrieval problems, improving representation ability, and high average accuracy

Pending Publication Date: 2022-03-08
GUANGDONG POLYTECHNIC NORMAL UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problems of unbalanced data and low accuracy of retrieval results in existing trademark retrieval methods in the background technology, the present invention provides an unbalanced trademark retrieval method based on deep hashing of aggregation similarity
This method combines the advantages of deep learning technology and hashing te

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
  • Unbalanced trademark retrieval method based on deep hash of aggregation similarity
  • Unbalanced trademark retrieval method based on deep hash of aggregation similarity
  • Unbalanced trademark retrieval method based on deep hash of aggregation similarity

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0060] An unbalanced trademark retrieval method based on deep hashing of aggregation similarity, specifically comprising the following steps:

[0061] S1. Construct the network architecture of the deep hash neural network based on aggregated similarity, and initialize the neural network parameters;

[0062] S2. Generate a trademark hash cluster center C={c through Hadamard matrix and good properties of Bern random sampling 1 ,c 2 ,...,c m},in, The trademark hash cluster center m is equal to the total number of trademark classes, and q is the number of digits of the hash code;

[0063] S3. Based on the trademark hash cluster center C preset in S2, construct a trademark semantic hash cluster center C'={c 1 ',c 2 ',...,c N ′}, N is the total number of samples in the trademark training set;

[0064] S4. For the training sample x of the trademark data set training set i , obtain the training set hash encoding set through the sparse hash encoding module;

[0065] S5. Calcu...

Embodiment 2

[0067] Based on Embodiment 1, this embodiment further explains the technical solution of an unbalanced trademark retrieval method based on deep hashing of aggregation similarity in the present invention.

[0068] An unbalanced trademark retrieval method based on deep hashing of aggregation similarity, specifically comprising the following steps:

[0069] S1. Construct the network architecture of the deep hash neural network based on aggregated similarity, and initialize the neural network parameters;

[0070] Such as figure 1 As shown, ASDHN, a deep hash network based on aggregated similarity, consists of a feature extractor based on convolutional neural network, a hash encoder based on aggregated similarity, and a classifier based on Hamming distance matching. The feature extractor selects the convolutional neural network architecture AlexNet as the backbone, such as figure 2 shown. The convolutional neural network architecture AlexNet consists of 7 weight layers, includi...

Embodiment 3

[0117] Based on the aggregation similarity-based deep hash network of embodiment 1 and embodiment 2 in the trademark retrieval method, this embodiment is respectively in the DrinkLogos-50 data set (can be found at github / xxx.com) and the public data set FlickerLogos-32 Experiments were carried out on the platform, and the current hashing methods with better effects in trademark retrieval applications were compared horizontally.

[0118] In the experiment, the hardware platform used is Ubuntu18.4 LTS system, 64GB memory, (NIVIDIA) GeForceRTX TM 2080 Ti graphics card*4; software environment is Python 3.7, Torchvision 0.5.0, Pytorch 1.4.

[0119] In the comparison process, the method of the present invention and the most advanced method of trademark hash retrieval are adopted respectively, and the backbone networks adopted are respectively AlexNet and ResNet. The network has been pretrained on ImageNet. Models employing the AlexNet backbone are marked with an asterisk.

[0120...

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 an unbalanced trademark retrieval method based on deep hash of aggregation similarity, which comprises the following steps of: 1, constructing a network architecture of a neural network, and initializing neural network parameters; 2, generating a trademark hash cluster center through good attributes of a Hadamard matrix and Bern random sampling; thirdly, constructing a trademark semantic hash cluster center based on the trademark hash cluster center; 4, for training samples of the trademark data set training set, obtaining a training set Hash coding set through a sparse Hash coding module; and 5, calculating argminLT loss and executing back propagation to optimize neural network parameters until the model converges, and generating a deep hash neural network based on aggregation similarity. Compared with the prior art, the method has the advantages that the advantages of the deep learning technology and the Hash technology are combined, similar trademarks are made to be close, different trademarks are made to be far away, therefore, the trademark training characterization capacity is improved, and the retrieval problem caused by imbalance between data classes is solved.

Description

technical field [0001] The invention relates to the field of computer technology, and more specifically, to an unbalanced trademark retrieval method based on deep hashing of aggregation similarity. Background technique [0002] As an important intellectual property right, trademark plays a pivotal role in social and economic development. How to quickly and effectively retrieve similar trademarks from a large number of trademarks will be very helpful to trademark examiners and trademark owner applicants. Therefore, it is of great practical significance to study accurate and fast trademark retrieval methods for huge trademark databases. [0003] Due to the late start of digitalization of the official patent database, the included trademark images are not complete. Therefore, like other datasets, official trademark databases suffer from data imbalance. If the training data is unbalanced, most existing learning algorithms will have a learning bias for the majority class, resu...

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
IPC IPC(8): G06F16/908G06K9/62G06N3/04G06N3/08
CPCG06F16/908G06N3/084G06N3/045G06F18/22G06F18/214G06F18/24
Inventor 林煜森戴青云雷方元杨驭涵
Owner GUANGDONG POLYTECHNIC NORMAL UNIV
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