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

A heuristic image scale normalization method based on attention mechanism

A technology of attention and normalization, applied in the fields of image recognition, convolutional neural network, and computer vision, can solve problems such as scaling and deformation, and achieve the effect of reducing waste and avoiding unfavorable changes

Active Publication Date: 2022-08-02
ZHEJIANG UNIV
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to improve the deficiencies of the current common image scale normalization methods and avoid the scaling and deformation problems caused by inconsistent image aspect ratios. The present invention proposes a heuristic image scale based on attention mechanism Normalization method

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
  • A heuristic image scale normalization method based on attention mechanism
  • A heuristic image scale normalization method based on attention mechanism
  • A heuristic image scale normalization method based on attention mechanism

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0047] An auxiliary classification network is trained on the ImageNet dataset. The auxiliary attention feature map generation network consists of multiple convolutional layers. The generation network is an encoder-decoder structure, including convolutional layers, ResBlock and transposed convolutional layers. The backbone classification network uses AlexNet with a high running rate.

[0048] By applying this network to our image to be normalized, we get the final normalized result. For specific examples, please refer to figure 2 .

[0049] We then experimented our normalization method on the image recognition task. The data set used in this experiment is the Chinese herbal medicine data set, which contains 15,485 images and a total of 556 types of herbal data.

[0050] We use three ways to normalize the image scale (direct scaling, random cropping, our method). Then use the normalized pictures to conduct experiments to obtain the classification accuracy of the model obtai...

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 heuristic image scale normalization algorithm based on an attention mechanism. The method first uses a dataset with a large amount of data to train a classification network with an attention mechanism, which can simultaneously output the attention feature map and the classification result of the image. The network is trained using the classification results during training. Taking advantage of the ability of attention feature maps to quantify the importance of pixels in an image, this method uses a sliding window to slide over the attention feature maps. When the value in the sliding window reaches the maximum value, the position corresponding to the sliding window is the position to be cropped of the original image. The present invention can transform the image to the specified size by this method without deformation. When the image needs to be cropped, the method will automatically select the appropriate area for cropping.

Description

technical field [0001] The invention relates to the fields of computer vision, image recognition and convolutional neural networks, in particular to a heuristic image scale normalization method based on an attention mechanism. Background technique [0002] At present, many image recognition related technologies use deep learning models. Such deep learning-based image recognition models usually require the input image size to have a fixed size. In general, when the size of the image to be input does not match the size required by the model, the image to be input will be directly scaled to match the size received by the model. However, when the aspect ratio of the image received by the model is inconsistent with the aspect ratio of the image to be recognized, scaling the image to be recognized will result in distortion. This will adversely affect the model's predictions. To solve the above problems, a heuristic image size normalization method based on attention mechanism is...

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): G06V10/25G06V10/764G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V10/25G06N3/048G06N3/045G06F18/213
Inventor 张引董建洲
Owner ZHEJIANG 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