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

Image classification method, training method of image classification model and equipment

A classification method and classification model technology, applied in neural learning methods, biological neural network models, character and pattern recognition, etc., can solve the problems of complex data, high cost of sample data labeling, and few data set labels, so as to improve the classification effect , good learning effect, pay attention to the effect of continuity

Pending Publication Date: 2021-10-01
山东幻科信息科技股份有限公司
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] However, due to the complexity of sample data in some application image fields (such as the medical image field), professionals are required to label, resulting in a huge cost of sample data labeling, and it is not easy to obtain a large amount of labeling data
In the image data set, there are few data set labels, and the performance of special classification tasks is poor

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, training method of image classification model and equipment
  • Image classification method, training method of image classification model and equipment
  • Image classification method, training method of image classification model and equipment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0059] This embodiment proposes an image classification method. figure 1 It is a flowchart of an image classification method provided by an embodiment of the present invention. Such as figure 1 As shown, the method includes S10-S30.

[0060] S10: Divide the image to be classified into multiple patches, and generate a patch vector corresponding to each patch; perform dimensionality reduction on each patch vector through a linear layer, and splice multiple dimensionality-reduced patch vectors to obtain a first sequence of vectors; A variable vector is embedded in the head of the first sequence vector to obtain a second sequence vector, wherein the variable vector has the same size as each dimension-reduced patch vector, and the variable vector corresponds to the multiple The patch that can best represent the features of the image to be classified among the patches.

[0061] Optionally, the image to be classified is divided into multiple patches and flattened into a sequence, ...

Embodiment 2

[0088] This embodiment provides a method for training an image classification model, which is used to train the image classification model formed by the image classification method described in Embodiment 1. figure 2 It is a flowchart of a method for training an image classification model provided by an embodiment of the present invention. Such as figure 2 As shown, the method includes steps S01-S06.

[0089] S01: Obtain a training data set D, wherein the training data set includes a labeled data set D l and unlabeled dataset D u , each training data is a training image, each labeled data d l label is d l the true class of y l .

[0090] S02: For each labeled data d l Perform a random data enhancement to obtain an enhanced labeled dataset For each unlabeled data d u Perform K random data enhancements to obtain K enhanced unlabeled datasets k=1,...,K, all d u of K The union of is recorded as Each unlabeled data d u of K Respectively input the image classif...

Embodiment 3

[0126] Figure 4 It is a flow chart of another image classification method provided by the embodiment of the present invention. The method is based on Transformer's semi-supervised algorithm to realize the network learning process of image classification, including training phase and prediction phase. Such as Figure 4 As shown, the method includes S1-S8.

[0127] S1: Predict pseudo-labels. Firstly, random data enhancement is performed on all unlabeled data, repeated K times, then the enhanced unlabeled data is input into the model for prediction, and K pseudo-labels are obtained, and finally the average operation is performed as the pseudo-label of unlabeled data.

[0128] S2: image block processing. The input image is divided into multiple patches and flattened into a sequence, and then the dimensionality reduction operation is performed through a learnable linear projection. Finally, a vector of the same size as the patch (referred to as "patch embedding vector") is em...

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 classification method, a training method of an image classification model and equipment. The image classification method comprises the following steps: segmenting a to-be-classified image into a plurality of patches, and performing dimensionality reduction on each patch vector through a linear layer to obtain a first sequence vector; embedding a variable vector into the head of the first sequence vector to obtain a second sequence vector; initializing a position coding vector of the second sequence vector, and embedding the initialized position coding vector into the second sequence vector to obtain an input vector; inputting the input vector into an encoder of a Transform model to obtain an encoding vector; taking a variable vector of the head part of the encoding vector as a feature vector of the to-be-classified image; and inputting the feature vector into a classifier of a Transform model to obtain a prediction category probability of the to-be-classified image. According to the invention, the classification effect of image classification is improved.

Description

technical field [0001] Embodiments of the present invention relate to the technical field of image classification, and in particular to an image classification method, a training method and equipment for an image classification model. Background technique [0002] In the field of image recognition and classification, deep learning within the scope of machine learning is an effective method, which has produced many excellent algorithms and networks, including the common convolutional neural network (Convolutional Neural Network, referred to as "CNN"), loop Four mainstream network structures: Recurrent Neural Network, Generative Adversarial Networks, and Reinforcement learning. [0003] However, due to the complex sample data in some application image fields (such as the medical image field), professionals are required to label, resulting in a huge cost of sample data labeling, and it is not easy to obtain a large amount of labeling data. There are few labels in the image dat...

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
IPC IPC(8): G06K9/62G06K9/46G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/213G06F18/2415
Inventor 张凯王瑞丰丁冬睿杨光远逯天斌王潇涵
Owner 山东幻科信息科技股份有限公司
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