Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

An image recognition method based on convolutional neural network

A convolutional neural network and image recognition technology, applied in the field of deep learning, can solve problems such as waste, difficult transfer of fully connected neural network gradients, and limitations on the number of network layers

Active Publication Date: 2022-07-19
JILIN JIANZHU UNIVERSITY
View PDF7 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Clearly this full join is wasteful, and the large number of parameters can quickly lead to overfitting
[0004] (2) The position information between pixels is not used
Trying to learn a large number of unimportant weights is bound to be very inefficient
[0005] (3) Limitation on the number of network layers
The more layers of the network, the stronger the expressive ability, but it is very difficult to train the deep artificial neural network through the gradient descent method, because the gradient of the fully connected neural network is difficult to pass more than 3 layers
Therefore, it is impossible to obtain a deep fully connected neural network, which limits its ability

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
  • An image recognition method based on convolutional neural network
  • An image recognition method based on convolutional neural network
  • An image recognition method based on convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0089] The training model in this example has a total of 40 epochs to update the learning rate. A larger learning rate is set at the beginning of the training, and the learning rate gradually decreases as the total system error decreases during the learning process. In one epoch training, the best weights are saved for later deployment of the neural network model. During the training process, the SGD stochastic gradient descent method is used to optimize the training system, and minibatch training is used to speed up the model convergence. After completing 40 epochs of training, the best weights in the training are saved, and in the model prediction, the saved best weights are directly called to initialize the model prediction parameters to start predicting pictures.

[0090] Before training starts, load the images to be trained, preprocess the training set, including image normalization, image channel unification, etc., and then build and train the model, that is, start forwar...

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 belongs to deep learning and image recognition technology, in particular to an image recognition method based on a convolutional neural network. The method includes: using a convolutional neural network for model training on the original image; inputting the image to be processed into the trained model to identify the image. The method of the invention accelerates the training of the neural network by adopting the GPU mode in the training process, adds Dropout regularization to the training model to optimize the system to prevent over-fitting during the training process, and at the same time performs atlases on the photos of the data set. expansion.

Description

technical field [0001] The invention belongs to a deep learning and image recognition technology, in particular to an image recognition method based on a convolutional neural network. Background technique [0002] Since Rumelhart and his colleagues developed learning algorithms in 1985, there has been a worldwide upsurge in exploring and researching neural networks. The development of artificial neural networks has penetrated into this research field, especially in the image classification technology of pattern recognition. The applications are gradually increasing, and there are many domestic and foreign researches on character recognition technology, license plate recognition technology, face recognition technology, various banknote recognition technology, seal recognition technology, and the recognition of some military targets. When the artificial neural network completes the image recognition task, there are mainly the following problems: [0003] (1) Too many paramete...

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): G06V10/764G06V10/774G06K9/62G06V10/82G06N3/04G06N3/08
CPCG06N3/084G06N3/045G06F18/214G06F18/241
Inventor 刘航白仞祥张玉红菅秀凯刘鸣泰
Owner JILIN JIANZHU UNIVERSITY
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com