Image recognition method based on optimized deep neural network model

A deep neural network and model technology, applied in the field of image recognition, can solve problems such as increased model noise, average gradient decrease of network parameters, and decreased training performance, achieving the effects of speeding up, improving image recognition performance, and improving accuracy

Pending Publication Date: 2020-08-25
INST OF MICROELECTRONICS CHINESE ACAD OF SCI
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0002] At present, when the deep learning of neural network is used for image recognition, it is of great practical application value to increase the parallel computing scale of the model. Identify issues with degrading training performance
And through further research, it is found that this decline in performance is accompanied by an increase in model noise and a decrease in the average gradient of network parameters
In the existing public literature, a training method that sets the learning rate based on the gradient within the model layer is proposed, that is, LARS, which greatly improves the performance of model training when the batch size is large. size) value continues to increase, the training performance will inevitably decline

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 recognition method based on optimized deep neural network model
  • Image recognition method based on optimized deep neural network model
  • Image recognition method based on optimized deep neural network model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0036] Preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of the application and together with the embodiments of the present invention are used to explain the principle of the present invention and are not intended to limit the scope of the present invention.

[0037] A specific embodiment of the present invention, such as figure 1 Shown, disclose a kind of image recognition method based on optimizing deep neural network model, comprise the following steps:

[0038] S1, select the data set of the image as the training sample, divide the data set into multiple sample sets according to the batch setting of the batch number, input the current sample set into a specific deep neural network model, and obtain the loss function value set corresponding to the current sample set ; Wherein, the specific deep neural network model is selected as a deep neural n...

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 recognition method based on an optimized deep neural network model, and belongs to the technical field of image recognition. The method comprises the following steps:S1, selecting a data set of an image as a training sample, dividing the data set into a plurality of sample sets according to batch setting of a batch number, and inputting a current sample set intoa specific deep neural network model to obtain a loss function value set corresponding to the current sample set, wherein the specific deep neural network model is used as a selected deep neural network; s2, updating network parameters of the selected deep neural network by discarding or forcibly zeroing training samples corresponding to a part of sets with smaller numerical values in the loss function value set; s3, iterating S1 and S2 for multiple times to obtain an optimized deep neural network model; and S4, inputting a to-be-identified image into the optimized deep neural network model toobtain an identified image. According to the method, the problem that the training performance of the deep neural network model under large-batch image data sets is greatly reduced is solved.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to an image recognition method based on an optimized deep neural network model. Background technique [0002] At present, when the deep learning of neural network is used for image recognition, it is of great practical application value to increase the parallel computing scale of the model. Identify issues with degrading training performance. And through further research, it is found that this performance decline is accompanied by an increase in model noise and a decrease in the average gradient of network parameters. In the existing public literature, a training method that sets the learning rate based on the gradient within the model layer is proposed, that is, LARS, which greatly improves the performance of model training when the batch size is large. size) value continues to increase, the training performance will inevitably decline. Contents of the invention [0...

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): G06N3/04G06N3/08
CPCG06N3/084G06N3/045
Inventor 高峰利钟汇才王师峥
Owner INST OF MICROELECTRONICS CHINESE ACAD OF SCI
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products