A target classification and recognition method based on multi-source domain fusion transfer learning

A technology of transfer learning and target classification, which is applied in character and pattern recognition, instruments, computer parts, etc., can solve the problem of insufficient accuracy of transfer learning in the single-source field, and achieve a reasonable evaluation reliability effect

Inactive Publication Date: 2019-02-15
NORTHWESTERN POLYTECHNICAL UNIV
View PDF0 Cites 10 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a target classification and recognition method based on multi-source domain fusion transfer learning, which uses weighted DS rules to fuse classification results from multiple source domains, and solves the problem that the correct rate of single-source domain transfer learning is not high enough

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 target classification and recognition method based on multi-source domain fusion transfer learning
  • A target classification and recognition method based on multi-source domain fusion transfer learning
  • A target classification and recognition method based on multi-source domain fusion transfer learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0067] Example 1: Assume that the number of target sample categories to be identified is 3, that is, the identification framework is Ω={ω 1 ,ω 2 ,ω 3}, there are three source domains and one target domain, and the initial classification results are obtained as three evidences (the first classification results) through the classic transfer learning algorithm, and the trust assignment of each evidence is taken as

[0068] m 1 (ω 1 )=0.5,m2 (ω 2 )=0.3,m 3 (ω3) = 0.2;

[0069] m 2 (ω 1 )=0.6,m 2 (ω 2 )=0.2,m(ω 3 ) = 0.2;

[0070] m 3 (ω 1 )=0.2,m 3 (ω 2 )=0.5,m 3 (ω 3 ) = 0.3;

[0071] For the convenience of illustration, the distances calculated before and after mapping are directly given as

[0072]

[0073]

[0074] According to the above two distances, the final distance is The discount coefficient (reliability of the first classification result) obtained from the estimated distance is

[0075] beta 1 =1,β 2 =0.6774,β 3 = 0.6862. The weighted fu...

Embodiment 2

[0078] Embodiment 2: Assume that the basic trust assignment (first classification result) obtained by transfer learning is

[0079] m 1 (ω 1 )=0.7,m 2 (ω 2 )=0.1,m 3 (ω 3 ) = 0.2;

[0080] m 2 (ω 1 )=0.2,m 2 (ω 2 )=0.2,m(ω 3 ) = 0.6;

[0081] m 3 (ω 1 )=0.2,m 3 (ω 2 )=0.5,m 3 (ω 3 ) = 0.3;

[0082] The rest of the parameters are the same as mentioned in Example 1. It can be seen from the observation that the three evidences at this time respectively support ω 1 ,ω 3 ,ω 2 , if only one source field is considered, one of the three evidences will be obtained. However, the categories supported by these three evidences are different. Obviously, when there is only one evidence, the classification results obtained are prone to errors, that is, the classification results of a single source field There is a high chance of error. Therefore, the trust assignment (final classification result) obtained by weighted fusion method is

[0083] m(ω 1 )=0.5774, m(ω 2 )=...

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 target classification and identification method based on multi-source domain fusion transfer learning. The method comprises the following steps: determining at least two source domains and target domains, and calculating a first distribution distance between each source domain and target domain. Matching each source domain and the target domain, classifying the sample inthe target domain according to each matching result, and obtaining a first classification result of the sample; Calculating a second distribution distance; Calculating the reliability of each matchingresult according to the first distribution distance and the second distribution distance; Calculating according to each first classification result and corresponding reliability, obtaining n second classification results of the sample on the basis of corresponding source domain; Fusing a plurality of second classification results to obtain a final classification result of the sample; The invention adopts weighted DS rules to fuse multiple second classification results, and solves the problem that the learning accuracy rate of single-source domain migration is not high enough.

Description

【Technical field】 [0001] The invention belongs to the technical field of target recognition, and in particular relates to a target classification and recognition method based on multi-source field fusion transfer learning. 【Background technique】 [0002] In the field of machine learning, traditional classification recognition algorithms usually require training samples and test samples to obey independent and identical distribution and have the same feature space. In practical applications, these two conditions are often difficult to meet at the same time. Transfer learning is a new machine learning framework. The conditional constraints of traditional machine learning are relaxed. Migration learning is to use training samples in other related fields (source field) to assist the classification and identification of samples in the target field. There are sufficient labeled training samples in the source field, but there are few or no labeled samples in the target field, and ...

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): G06K9/62
CPCG06F18/25G06F18/214
Inventor 刘准钆黄林庆潘泉何友
Owner NORTHWESTERN POLYTECHNICAL 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