Image recognition method and device based on convolutional neural network

A technology of convolutional neural network and image recognition device, which is applied in the field of image recognition based on convolutional neural network, can solve the problems of low image recognition efficiency and achieve the effect of improving image recognition efficiency

Inactive Publication Date: 2018-04-20
北京一维大成科技有限公司
View PDF5 Cites 24 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] In view of this, the embodiment of the present invention provides an image recognition method and device based on a convolutional neural network, which solves the problem of low image recognition efficiency existing in existing image recognition technologies

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 recognition method and device based on convolutional neural network
  • Image recognition method and device based on convolutional neural network
  • Image recognition method and device based on convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0028] figure 1 It is a flowchart of an image recognition method in Embodiment 1 of the present invention. This embodiment is applicable to image recognition in various scenarios, such as face recognition, license plate recognition and biomedical image recognition. This method can be implemented by an image recognition device For execution, the image recognition device may be integrated in CPU and / or GPU. Such as figure 1 As shown, the image recognition method specifically includes the following steps:

[0029] S110. Extract a quantified quality value of the image to be recognized based on the image quality prediction branch in the convolutional neural network; wherein the quantified quality value is used to represent the degree of recognition of the image to be recognized.

[0030] The convolutional neural network described in this embodiment is obtained through training, and the convolutional neural network includes two parts: an image quality prediction branch and an imag...

Embodiment 2

[0046] figure 2 It is a flow chart of an image recognition method provided by Embodiment 2 of the present invention. This embodiment adds a training process to a convolutional neural network on the basis of the above embodiments, and specifically includes the following steps:

[0047] S210. Acquire at least three identity data sets as training data sets, and use the training data sets as an input of a base model. Wherein, the training data set includes at least two identity data sets representing the same image content, and at least two identity data sets representing different image content; the identity data set includes multiple training images representing the same image content.

[0048]Among them, the images represented by the images in each identity data set have the same image content. For example, the image content represented by the images in the A identity data set is the face image of user A, and the image content represented by the images in the B identity data s...

Embodiment 3

[0074] On the basis of the above-mentioned embodiments, this embodiment proposes a preferred embodiment based on the scene of face image recognition, which specifically includes the following steps:

[0075] For all face images I in the training dataset i (i=1,2,...N train ) for preprocessing. The preprocessing operation includes face detection and key point extraction, extracting the face image and recording the key points of the face, and then aligning each input image according to the key points, and uniformly classifying them into 245*245 scale. Record the face image after processing as I i (i=1,2,...N train ).

[0076] When inputting into the base model for training, a batch of training data sets with a specific composition are first selected for each round of training. The specific method is to randomly select three identity data sets as a set of training data sets, respectively S a , S p and S n . Among them, S a Represents the identity data set of user a, S ...

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 image recognition method and device based on a convolutional neural network. The image recognition method comprises the steps that the quality quantification value of an image to be recognized is extracted based on an image quality prediction branch in the convolutional neural network, wherein the quality quantification value is used to represent the degree of recognition of the image to be recognized; the quality quantification value is compared with a quality discrimination preset threshold; if the quality quantification value is greater than or equal to the quality discrimination preset threshold, the image to be recognized is recognized based on an image feature extracting branch in the convolutional neural network; and if the quality quantification value isless than the quality discrimination threshold, the image to be recognized is not recognized. According to the invention, the problem of low image recognition efficiency in the existing image recognition technology is solved; the quality of the image to be recognized is discriminated; the to-be-recognized image whose quality quantification value is less than the quality discrimination preset threshold is rejected; computing resources required by recognition can be greatly reduced; and the image recognition efficiency and accuracy can be improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an image recognition method and device based on a convolutional neural network. Background technique [0002] Image recognition technology is an important field of artificial intelligence. It refers to the technology of object recognition on images to identify targets and objects in various patterns. [0003] In the prior art, the image recognition technology simply preprocesses the image data, inputs it into the convolutional neural network for training to obtain the weight of the network, and calculates the feature vector according to the weight of the trained network, and obtains the image recognition result by processing the feature vector. result. [0004] However, in the prior art, there is a large difference in image quality between image data involved in training. Some low-quality images, such as low resolution, fuzzy, too strong or too dark lighting, large deflection o...

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): G06K9/46G06K9/62G06K9/00G06N3/04
CPCG06V40/168G06V40/193G06V10/462G06N3/045G06F18/214
Inventor 张静普
Owner 北京一维大成科技有限公司
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