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

Method for optimizing exception detection model based on deep learning

An anomaly detection and deep learning technology, applied in the field of machine learning, can solve the problems of multi-hours and labor cost preprocessing, high preprocessing cost, achieve the effect of achieving model generalization performance and solving the problem of imbalance between classes

Active Publication Date: 2019-08-27
BEIJING UNIV OF POSTS & TELECOMM
View PDF10 Cites 8 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem that in the case of sustainable acquisition of samples, the cost of data preprocessing in the prior art is high, and it takes more man-hours and labor costs to preprocess the unbalanced samples of the sample library, the present invention proposes a depth-based Sample library expansion method for learning anomaly detection model

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
  • Method for optimizing exception detection model based on deep learning
  • Method for optimizing exception detection model based on deep learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0020] Example 1: Initial model training

[0021] (1) Basic sample preparation: Collect image data of production products in industrial production, with 1000 as the basic quantity.

[0022] (2) Training model: Based on the deep learning anomaly detection algorithm, train the anomaly detection model.

Embodiment 2

[0023] Embodiment 2: real-time detection

[0024] (1) Model detection: While inputting sample data in real time, model detection is performed, and the classification probability of the output sample is output.

[0025] (2) Calculate the effective sample probability interval: let the original data set, the number of positive samples be N P , the number of negative samples is N N , the number of predicted samples is P, and the number of wrong samples is P F , σ is the sigmoid function, so that the probability interval of the effective sample, that is, the fuzzy interval predicted by the model is: where parameters Defined as 0.5, the parameter β is defined as 0.5. The influencing factors in the formula are: the proportional coefficient of positive and negative samples in the original data set, when the class is unbalanced, the interval for judging negative samples should be increased; the accuracy of model detection, when the accuracy rate becomes higher, the range of fuzzy...

Embodiment 3

[0027] Embodiment 3: Judging whether the balance condition of the sample library is met

[0028] (1) Determine whether the sample library meets the requirements: the number of positive and negative samples satisfies N P N .

[0029] (2) Loop processing: When the sample library does not meet the requirements, the loop is started, model training is performed on the new sample library, and the anomaly detection model is iteratively optimized.

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 method for optimizing an abnormal detection model based on deep learning, and the method mainly comprises the steps: collecting image data of a product in industrial production as a basis, and forming an original sample set; constructing a deep convolutional network, and predicting a picture sample in production by using a neural network model; marking the prediction sample, and calculating an effective sample probability interval; and inputting the effective samples obtained through calculation into a sample library, and performing model training iteration until thesample library is balanced between classes. According to the method, effective samples in the prediction samples are screened, inter-class balance of the sample library is achieved, the generalizationperformance of the model is optimized, and the prediction effect is obviously improved.

Description

technical field [0001] The invention relates to the field of machine learning, in particular to an iterative optimization method for an abnormality detection model based on deep learning. Background technique [0002] With the advancement of science and technology, industrial product equipment is becoming more and more complex and sophisticated, the increase of system components and influencing factors, and the complexity of operating environment and working conditions make product anomaly detection face great challenges. The deep learning model has outstanding advantages in feature learning. The learned features can better describe the rich information of the data and improve the classification performance. It has been widely used in anomaly detection. However, in the actual industrial process, compared with the normal industrial process, there are too few defective images in the collected product image data, adding a large number of normal samples and a small number of def...

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): G06F16/50G06N3/08G06N3/04
CPCG06N3/08G06N3/045
Inventor 张永军沈涛闫思宇文韩
Owner BEIJING UNIV OF POSTS & TELECOMM
Features
  • Generate Ideas
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2025 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com