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

Defect detection device and method based on distillation learning mechanism, and storage medium

A defect detection and learning mechanism technology, applied in the field of anomaly detection, can solve problems such as difficulty in collection, achieve the effects of simple deployment, optimized expression capabilities, and increased differences

Pending Publication Date: 2021-10-01
ZHEJIANG LINYAN PRECISION TECH CO LTD
View PDF15 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] At present, most industrial product defect detection based on deep learning requires a large amount of product image data, which needs to contain defective data and non-defective data, so as to learn a model with high accuracy, but there is no such thing in the actual scene. It is easier to obtain image data of defective industrial products, but the image data of defective industrial products appears randomly and is more difficult to collect, so the supervised learning method is not suitable for this scenario

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
  • Defect detection device and method based on distillation learning mechanism, and storage medium
  • Defect detection device and method based on distillation learning mechanism, and storage medium
  • Defect detection device and method based on distillation learning mechanism, and storage medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0061] A defect detection device based on a distillation learning mechanism, including an acquisition module, a training module, and a detection module; the acquisition module is used to collect flawless industrial product image samples and form a training set; the training module is used to use the training set Train the network model, the network model includes a teacher model and a student model with the same backbone network structure, adopt the teacher model to guide the student model, and finally obtain an optimized network model; the detection module is used to input the image to be tested into the optimized network model and Output the detection result.

[0062] Such as figure 1 As shown, the complexity of the network layer of the teacher model is higher than the complexity of the network layer of the student model, and the middle layer between the teacher model and the student model is added with a feature information grafting module, which is used for synchronizing t...

Embodiment 2

[0066] This embodiment is optimized on the basis of embodiment 1, such as figure 2 As shown, the feature information grafting module includes a feature information difference value calculation part and a grafted feature information calculation part, and the feature information difference value calculation part is used to carry out the feature information of the residual combination blocks of the teacher model and the student model at the same level The difference value calculation, such as image 3 As shown, when the difference value is greater than the threshold value, enter the grafting feature information calculation part, fuse the teacher model feature information with the student model feature information, and then replace the original student model feature information, when the difference value is less than or equal to the threshold value , the feature information of the student model will not be replaced.

[0067] Further, as image 3 As shown, after the feature info...

Embodiment 3

[0084] This embodiment is optimized on the basis of embodiment 1 or 2, as figure 1 As shown, the backbone networks of the teacher model and the student model respectively include a convolution layer, a batch normalization layer, an activation function layer, and several residual modules arranged sequentially from front to back. The activation function layer uses parameters to modify the linear unit layer.

[0085]Further, the teacher model adopts the ResNet101 structure, the student model adopts the ResNet20 structure, the teacher model and the student model respectively include 4 residual combination blocks, and the number of residual modules contained in the 4 residual combination blocks of the teacher model The order is 6, 12, 24, 6, and the number of residual modules contained in the 4 residual combination blocks of the student model is 1, 2, 4, 1 in order.

[0086] Other parts of this embodiment are the same as those of Embodiment 1 or 2 above, so details are not repeate...

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 flaw detection device and method based on a distillation learning mechanism, and a storage medium. A teacher model and a student model are constructed according to design, wherein the teacher model and the student model are consistent in structure, but the network layer complexity of the teacher model is greater than that of the student model; and then a feature information grafting module is added to the middle layer to accelerate the learning progress and detection precision of the student model; and finally, a loss function is used to calculate a loss value to optimize model parameters. The feature information grafting module selects whether to graft the calculated fusion feature information into the student model by calculating a feature information difference value between the teacher model and the student model, so that the ability of the student model to simulate the teacher model is improved. According to the invention, a teacher model and a student model are built by using a distillation learning mechanism, and then a feature information grafting module is introduced into a double-branch network intermediate layer. The method can optimize the expression ability of the student model, accelerate the learning process of the student model, and improve the detection precision of the student model.

Description

technical field [0001] The invention belongs to the technical field of anomaly detection, and in particular relates to a defect detection device, method and storage medium based on a distillation learning mechanism. Background technique [0002] With the prosperity and development of the national economy, the rapid development of the manufacturing industry has been promoted, and the automatic production technology of industrial products has become a trend. Industrial products will more or less have defects during the production process, such as printed circuit board bad points, textile appearance defects, electronic screens and other inevitable surface defects. Timely detection and removal of these unqualified products can ensure The quality of industrial products can also greatly improve production efficiency. [0003] Industrial product defect detection methods are mainly divided into traditional methods and artificial intelligence methods. There are also two traditional...

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): G06T7/00G06K9/62G06N3/04G06N3/08
CPCG06T7/0004G06N3/08G06T2207/20081G06T2207/20084G06T2207/30108G06N3/045G06F18/22G06F18/214Y02P90/30
Inventor 张晓武陈斌
Owner ZHEJIANG LINYAN PRECISION TECH CO LTD
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