Unlock instant, AI-driven research and patent intelligence for your innovation.

Image classification method and system based on lightweight Non-Local

A classification method and a lightweight technology, applied in the direction of neural learning methods, instruments, biological neural network models, etc., can solve problems such as high calculation load, calculation load increase, and large calculation load, so as to achieve compatibility and reduce computational complexity , Simplify the effect of the long connection relationship acquisition method

Pending Publication Date: 2022-05-13
NANJING UNIV OF SCI & TECH
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] However, while Non-Local brings powerful feature expression capabilities, it also has many defects, such as bringing a high amount of calculation
Therefore, the image classification method using the Non-Local method also has the problem of a large amount of calculation. Based on this, the present invention provides a light-weight Non-Local-based image classification method and system, which can not only enhance the characteristics of pixels Expressive ability to improve classification accuracy and avoid a large increase in calculation

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 classification method and system based on lightweight Non-Local
  • Image classification method and system based on lightweight Non-Local
  • Image classification method and system based on lightweight Non-Local

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0039] This embodiment provides an image classification method based on lightweight Non-Local, please refer to figure 1 , the method includes:

[0040] S1. Input the sample image into a lightweight network, and obtain abstract features after each bottleneck layer of the lightweight network, and the lightweight network includes ShufflenetV2.

[0041] Set the model processing task as an image classification task in computer vision, and divide the image into a training set and a test set according to the general guidelines in image classification tasks, such as CIFAR100, ImageNet. The backbone network used is set to be a lightweight network with limited channels, such as ShufflenetV2. After the image enters the network as input, abstract features are obtained after each bottleneck layer Among them, C, H and W are respectively the number of channels, height and width of feature X, and the present invention enhances this abstract feature.

[0042] S2. Divide the abstract featur...

Embodiment 2

[0079] This embodiment provides a lightweight Non-Local-based image classification system, the system comprising:

[0080] The abstract feature acquisition module M1 is used to input the sample image into the lightweight network, and obtain abstract features after each bottleneck layer of the lightweight network, and the lightweight network includes ShufflenetV2;

[0081] A grouping module M2, configured to divide the abstract features into G groups along the channel dimension;

[0082] The spatial weight acquisition module M3 is used to obtain a spatial weight by convolution for the abstract features of each group, and the spatial weight is the global spatial attention information for the current group;

[0083] The unary long connection feature acquisition module M4 is used to use the global spatial attention information of the current group to perform weighted magnitude conversion on the pixels of the group to obtain the unary long connection feature of the current group; ...

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 relates to an image classification method and system based on lightweight Non-Local, and belongs to the field of image classification. According to the method, the abstract features behind a bottleneck layer are grouped by utilizing the grouping implicit clustering characteristic of the lightweight network, so that an original Non-Local long connection relation acquisition mechanism is simplified, and the calculation complexity is reduced. The unary long connection relation and the point-to-long connection relation are utilized to extract the features, the calculation complexity is reduced, the long connection performance originally brought by Non-Local is reserved, and compatibility of the calculation complexity and Non-Local long connection is achieved. And predicting the type of the image by using the extracted features, calculating the cross entropy loss of the predicted image type and the actual image type, training a model according to the cross entropy loss to obtain a trained model, and carrying out image classification by using the trained model to obtain an image classification result. The image classification accuracy can be improved, and the calculated amount can be prevented from being greatly increased.

Description

technical field [0001] The invention relates to the field of image classification, in particular to an image classification method and system based on lightweight Non-Local. Background technique [0002] The Non-Local self-attention mechanism was proposed by Wang Xiaolong and He Kaiming at CVPR2018 (Computer Vision Summit). The Non-Local method uses the feature similarity between image pixels to establish a long connection relationship between pixels to enhance the feature expression ability of each pixel in the image. This relationship does not have the local limitations of convolution in neural networks, that is, the Non-Local mechanism can explore long connection relationships regardless of the distance between pixels. In other words, Non-Local methods have domain capabilities that cannot be achieved by convolutions. [0003] Since the non-local self-attention mechanism was proposed, it has been applied in the classic deep learning convolutional neural network, and it h...

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): G06K9/62G06V10/764G06V10/80G06V10/774G06V10/82G06N3/04G06N3/08
CPCG06N3/084G06N3/045G06F18/24G06F18/253G06F18/214
Inventor 孙运莲庄程
Owner NANJING UNIV OF SCI & TECH