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

Lightweight image classification method, system and device based on model pruning

A classification method and lightweight technology, applied in the field of computer vision and deep learning, can solve the problems of unmeasured features and classification task correlation, and achieve the effect of reducing the loss of image classification accuracy, good feature representation ability, and easy deployment and application

Active Publication Date: 2021-04-16
HUAZHONG NORMAL UNIV
View PDF6 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, information entropy can only measure the amount of information contained in feature channel extraction features, and cannot measure the correlation between the extracted features and classification tasks.
In addition, although this method can compress the convolutional neural network well, it cannot avoid the loss of accuracy caused by pruning.

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
  • Lightweight image classification method, system and device based on model pruning
  • Lightweight image classification method, system and device based on model pruning
  • Lightweight image classification method, system and device based on model pruning

Examples

Experimental program
Comparison scheme
Effect test

example

[0074] This example is tested on the standard dataset CIFAR-10. CIFAR-10 is an image classification data set. All images are three-channel color images with a size of 32×32, containing 60,000 pictures, of which 50,000 are for the training set and 10,000 are for the verification set. The deep neural network used in the experiment in this example is AlexNet, VGG16 and residual network ResNet.

[0075] In order to clearly see the changes before and after compression, this example first uses different convolutional neural networks to train a full-precision 32-bit model. In the compression process, the pre-trained 32-bit model is used as the initialization of the quantized model. In the process of pre-training and pruning, the data pre-processing uses the method of data enhancement, filling 0 on the original 32*32 image boundary and expanding it to a 36*36 image, and then randomly cropping it into a 32*32 image, and then Randomly flip left and right.

[0076] Table 1 Optimizatio...

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 lightweight image classification method, system and device based on model pruning. The method comprises the following steps: calculating the importance of a mutual information evaluation channel; performing sorting and grouping according to the importance of the channels; setting an attention regularization loss function according to grouping and optimizing the model; pruning unimportant channels; and optimizing the model again, and recovering the precision of the model. According to the method, system and device, the compression effect on the deep neural network is obvious, the precision loss caused by pruning is reduced while the storage and calculation consumption of a large-scale deep network model is reduced, and the method, system and device can be better applied to mobile terminal equipment for image classification.

Description

technical field [0001] The invention belongs to the technical field of computer vision and deep learning, and more specifically relates to a lightweight image classification method, system and equipment based on model pruning. Background technique [0002] In recent years, deep convolutional neural networks have been widely used in machine learning fields such as computer vision due to their good performance. In order to improve the performance of deep convolutional neural network in image classification, an extremely deep network structure is often used, that is, a large number of convolutional layers are cascaded, which leads to a huge number of parameters in the deep neural network model, which requires a large amount of calculation and storage. resources, which limits its deployment and application in edge devices such as smart mobile terminals. [0003] In order to solve this problem, an effective solution is to use a lightweight neural network model, that is, to prune...

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): G06K9/62G06N3/04G06N3/08
Inventor 陈靓影徐如意杨宗凯柏宝
Owner HUAZHONG NORMAL UNIV
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