Small sample target recognition method based on feature subspace generative adversarial network

A feature subspace and target recognition technology, applied in the field of deep learning, can solve problems such as the disadvantage of small sample target recognition

Active Publication Date: 2021-07-20
HANGZHOU DIANZI UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Although the target recognition method based on deep learning has high recognition accuracy, it has a relatively weak point in the problem of small sample target recognition.

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
  • Small sample target recognition method based on feature subspace generative adversarial network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0022] The present invention is further analyzed below in conjunction with specific examples.

[0023] Such as figure 1 As shown in , this experiment uses a set of collected ship target images as the training sample data set. The specific steps in the small sample target recognition task based on the feature subspace generation confrontation network are as follows:

[0024] Step (1), Obtain the features used to train the Generative Adversarial Network

[0025] 1.1 Use the classification network to perform rough training on image sample data.

[0026] Use the convolutional neural network to perform rough training on the image sample data. After training, the weight parameters and bias parameters are W and B respectively;

[0027] 1.2 Use W and B two sets of parameters to map the sample to the feature subspace to obtain feature data.

[0028] Assuming that there are n images in the sample data, the n image sample data is set A, and each image sample is A k (where 1

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 small-sample target recognition method based on feature subspace generation confrontation network; the invention first obtains the features used for training generation confrontation network, then uses the reorganized feature data to train generation confrontation network, and then uses the trained The generated confrontation network generates new features to expand the small sample data, and finally uses the expanded data to train the target recognition network. The key of the present invention is to imitate the enhancement strategy for underwater images degraded by different factors by using the data-driven characteristics of the generative adversarial network and strong imitation ability. The multi-branch structure of attribute branch and discriminative branch is used to enhance the comprehensiveness and robustness of learning.

Description

technical field [0001] The invention belongs to the field of deep learning and relates to a small-sample target recognition method for generating an adversarial network in a feature subspace. Background technique [0002] Few-shot object recognition is a problem often encountered in the field of deep learning. Although the target recognition method based on deep learning has high recognition accuracy, it has a relatively weak point in the problem of small sample target recognition. Therefore, it is necessary to use the generative confrontation network to expand the small sample data. [0003] Generative Adversarial Network (GAN) is a network structure that is trained against a Generative Network (G) and a Discriminative Network (D). The purpose of the G network is to make the synthesized image as close as possible to the real image, making it indistinguishable for the D network, which aims to distinguish the real image from the synthetic image. GAN has been successfully a...

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): G06K9/00G06N3/04G06N3/08
CPCG06N3/08G06V20/20G06V2201/07G06N3/045
Inventor 陈华杰侯新雨姚勤炜张杰豪韦玉谭刘栋
Owner HANGZHOU DIANZI 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