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

Residual network model based on cross-stage local feature fusion strategy and training method of model

A technology of local features and network models, applied in biological neural network models, neural learning methods, computer components, etc., can solve problems such as network degradation, achieve the effect of improving accuracy, solving network degradation problems, and enhancing learning ability

Pending Publication Date: 2022-07-29
浙大宁波理工学院
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Therefore, redundant network layer learning causes network degradation

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
  • Residual network model based on cross-stage local feature fusion strategy and training method of model
  • Residual network model based on cross-stage local feature fusion strategy and training method of model
  • Residual network model based on cross-stage local feature fusion strategy and training method of model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0055] Assuming that there is already an optimal model for a research task that can just make the classification task 100% accurate, its network layer number is 10 layers. When we build the network model, this optimal model exists but is unknown, so we may set the number of network layers of 22 layers, then in fact, 12 layers of the designed network structure are redundant and useless. Unless the 12 layers become an identity map during model training, that is, the output through these 12 layers is exactly the same as the input, it is difficult for us to guarantee that the performance of the 22-layer network model is as good as the optimal 10-layer network model. Therefore, redundant network layer learning causes network degradation. In practical research, on the one hand, it is necessary to avoid network degradation caused by too many network layers, and on the other hand, it is necessary to deepen the network, make full use of processor computing power, and improve network pe...

Embodiment 2

[0093] The present invention also proposes a training method for a residual network model, which is used to train the residual network model based on the cross-stage local feature fusion strategy according to any one of claims 1 to 5. Before predicting the tool wear state, the residual network model of the local feature fusion strategy also includes training the model, and the training method includes:

[0094] acquiring a training data set, the training data set includes a tool monitoring signal with a training label, and the training label is the tool wear amount at the monitoring moment corresponding to the tool monitoring signal;

[0095] A residual network model based on a cross-stage local feature fusion strategy is trained using the training dataset.

[0096] The final output of the model is the same as the training label of the sample point (tool monitoring signal). At present, the training label commonly used in tool wear monitoring research is to encode the tool wea...

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 residual network model based on a cross-stage local feature fusion strategy, which relates to the field of network models and comprises a residual block, and a left operation channel, a right operation channel and a merging unit are arranged in the residual block. The residual error feature merging module is used for merging the output residual error left feature vector and the residual error right feature vector to obtain a residual error feature merging vector; the system further comprises a local residual module which comprises at least one local residual block, the local residual block is used for splitting the merged vector to obtain a right feature vector and a left feature vector, and residual operation is carried out on the right feature vector through a right gradient operation channel in the local residual block to obtain a left feature vector; according to the method, the local residual block is arranged, so that paths of network branches are increased, and the problems that the number of network layers in a deep residual network is deepened, so that the number of network parameters is increased, and required calculation is also exponentially increased are solved.

Description

technical field [0001] The invention relates to the field of residual network models, in particular to a residual network model based on a cross-stage local feature fusion strategy and a training method for the model. Background technique [0002] In order to accurately predict the state of tool wear in CNC machine tools through the collected tool monitoring signals, it is necessary to establish a neural network model for predicting the state of tool wear. [0003] Compared with traditional machine learning methods, convolutional neural network transforms linearly inseparable problems into linearly separable problems based on deep networks, and can adaptively extract optimal features and establish models, which overcomes the need for prior experience and steps to manually extract features. Complex shortages. Compared with shallow learning, deep networks have more network layers and stronger data analysis capabilities. But this does not mean that we can infinitely increase ...

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): G06K9/00G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/048G06N3/045G06F2218/08G06F18/253
Inventor 白剑宇崔乾东白昊天文世挺杨劲秋
Owner 浙大宁波理工学院
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