Image classifier-oriented metamorphic test method and system

A technology of image classifier and metamorphosis test, which is applied in software testing/debugging, instrument, character and pattern recognition, etc.

Pending Publication Date: 2021-06-29
ARMY ENG UNIV OF PLA
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Benefits of technology

This patented method allows for accurate testing of neural networks by creating an artificially constructed model that can be used with images from different sources or domains without being affected by any changes made during training process.

Problems solved by technology

Technological Problem addressed in this patented technical solution described in the patents relates to improving the accuracy of predictive analysis systems like neural networks. Current methods have limitations due to their complexity and high cost associated with developing new models. To address these issues, researches aimed towards exploring alternative ways to verify accurate matches against previously-observed patterns.

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 classifier-oriented metamorphic test method and system
  • Image classifier-oriented metamorphic test method and system
  • Image classifier-oriented metamorphic test method and system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach

[0046] Embodiment: the metamorphic relationship test determination method of image classifier (the method flow process of the embodiment of the present invention is as follows figure 1 shown), including:

[0047] 1) Combining the collected original training set S and the original test set T with the metamorphic relationship, the derived training set S' and the derived test set T' can be obtained;

[0048] 2) Secondly, the original training set S and the derivative training set S′ are respectively trained on the already written machine learning image classification algorithm A, and the image classifier AT (namely the first image classifier) ​​and the image classifier AT′ ( i.e. the second image classifier);

[0049] 3) Test the original test set T and the derived test set T' on the image classifier AT and the image classifier AT' respectively;

[0050] 4) Judging the test results of the two classifiers O AT with O AT 'Whether it satisfies the metamorphosis relationship, so ...

Embodiment 1

[0054] Embodiment one: the metamorphosis relationship test determination method of image classifier, present embodiment is the metamorphosis test facing SVM image classifier, comprises the following steps:

[0055] First, the SVM algorithm is selected, and the data set uses the digits handwritten data set, which divides handwritten digital images into categories from 0 to 9. The machine learning algorithm is implemented based on the official Scikit-learn, a popular machine learning library. The algorithm can be divided into realization based on linear kernel function or nonlinear kernel function.

[0056] Then the metamorphosis relation is constructed for both the linear kernel function SVM algorithm and the nonlinear kernel function SVM algorithm. For training, the SVM algorithm takes labeled data of handwritten digits and learns how intrinsic properties in the training data separate the categories to derive decision boundaries. Each marker data is an image of 8 pixels by 8...

Embodiment 2

[0080] Embodiment two: the metamorphosis relation test determination method of image classifier, the present embodiment is the metamorphic test method for VGG16 image classifier, comprises the following steps:

[0081] The structure of the transformation relationship is as follows:

[0082] The second image classifier chosen is a deep learning image classifier called VGG16. The emergence of the VGG16 model has made breakthrough achievements in convolutional neural networks. The model can complete the task of image classification very well, and achieved excellent results in the 2014 ILSVRC competition. This embodiment implements VGG16 based on the tensorflow framework. The VGG16 image classifier is trained and tested on the labeled data, and the CIFAR-10 data set is selected. Each data instance of this dataset is a 32*32 pixel color image, that is, each instance has 32*32*3 feature numbers and has been divided into 10 mutually exclusive classes. Figure 6 A visualization of...

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 classifier-oriented metamorphic test method and system. The method comprises the following steps: respectively combining an original training set and an original test set with a pre-constructed metamorphic relationship to obtain a derivative training set and a derivative test set; according to the original training set, the original test set, the derivative training set and the derivative test set, carrying out correctness judgment on a pre-constructed metamorphic relation; and when the judgment result is that the metamorphic relation is correct, obtaining a variation image classifier by executing a variation operator, and performing effectiveness judgment on the correct metamorphic relation by using the original training set, the derivative training set, the derivative training set and the derivative test set for the image classifier. According to the invention, the correctness and effectiveness of the pre-constructed metamorphic relation facing the machine learning image classifier are judged, so that the problem of testing and judging the machine learning image classifier by using a metamorphic testing technology is solved.

Description

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

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
Owner ARMY ENG UNIV OF PLA
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

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap