Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
IMAGE CLASSIFICATION METHOD FOR EQUIVARIANT CONVOLUTIONAL NETWORK MODEL BASED ON PARTIAL DIFFERENTIAL OPERATOR
Document Type and Number:
WIPO Patent Application WO/2021/184466
Kind Code:
A1
Abstract:
An image classification method for an equivariant convolutional network model based on a partial differential operator. For an input layer and an intermediate layer of a convolutional network model, an equivariant convolution of the input layer and an equivariant convolution of the intermediate layer are respectively designed on the basis of a partial differential operator, and an equivariant convolutional network model PDO-eConv is constructed and performed model training; an input of the model PDO-eConv is image data, and an output of the model PDO-eConv is the predictive classification of an image, so that efficient image classification and recognition visual analysis is achieved. The method can provide a better parameter sharing mechanism, and achieve a lower image classification error rate.

Inventors:
LIN ZHOUCHEN (CN)
SHEN ZHENGYANG (CN)
HE LINGSHEN (CN)
Application Number:
PCT/CN2020/084650
Publication Date:
September 23, 2021
Filing Date:
April 14, 2020
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
UNIV BEIJING (CN)
International Classes:
G06K9/62; G06K9/54; G06N3/04; G06N3/08
Domestic Patent References:
WO2017142397A12017-08-24
Foreign References:
CN104517122A2015-04-15
CN108764289A2018-11-06
EP2869239A22015-05-06
EP2911111A22015-08-26
Other References:
SANDER DIELEMAN , JEFFREY DE FAUW , KORAY KAVUKCUOGLU: "Exploiting Cyclic Symmetry in Convolutional Neural Networks", ARXIV.ORG, 8 February 2016 (2016-02-08), pages 1 - 10, XP080682215
Attorney, Agent or Firm:
BEIJING WANXIANGXINYUE INTELLECTUAL PROPERTY OFFICE (CN)
Download PDF: