计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (17): 69-77.DOI: 10.3778/j.issn.1002-8331.1906-0242

• 理论与研发 • 上一篇    下一篇

机器学习图像分类程序的蜕变测试框架

刘佳洛,姚奕,黄松,惠战伟,陈强,寇大磊,张仲伟   

  1. 1.陆军工程大学 指挥控制工程学院,南京 210007
    2.中国人民解放军68023部队
    3.中国人民解放军73671部队
  • 出版日期:2020-09-01 发布日期:2020-08-31

Metamorphic Testing Framework for Machine Learning Image Classification Program

LIU Jialuo, YAO Yi, HUANG Song, HUI Zhanwei, CHEN Qiang, KOU Dalei, ZHANG Zhongwei   

  1. 1.College of Command and Control Engineering, Army Engineering University of PLA, Nanjing 210007, China
    2.Unit 68023 of PLA, China
    3.Unit 73671 of PLA, China
  • Online:2020-09-01 Published:2020-08-31

摘要:

机器学习在计算机视觉、语音识别和自然语言处理等实际应用中已经取得了显著的成功。图像分类作为计算机视觉的一个主要分支。不久的将来,许多的图像分类程序会以机器学习的方式呈现。然而,由于机器学习图像分类程序的测试面临着测试预言难题,这使得在测试的过程中将需要大量的人力及物力。为了缓解测试预言难题,使用了蜕变测试技术。为了规范测试流程、提高测试效率,提出了一种适用于机器学习图像分类程序的蜕变测试框架。并且通过测试基于SVM和VGGNet图像分类程序,验证了该测试框架的合理性和有效性。

关键词: 机器学习, 测试判定问题, 蜕变测试, 蜕变关系

Abstract:

Machine learning has achieved remarkable success in practical applications such as computer vision, speech recognition, and natural language processing. Image classification is a major branch of computer vision. In the near future, many image classification programs will be presented in machine learning. However, since the testing of machine learning image classification programs faces test oracle problems, it will require a lot of manpower and material resources in the testing process. In this paper, in order to alleviate the testing oracle problem, it uses the Metamorphic Testing(MT) technology. At the same time, in order to standardize the testing process and improve the testing efficiency, a metamorphic testing framework suitable for machine learning image classification program is proposed. Through testing SVM and VGGNet image classification programs, the rationality and effectiveness of the testing framework are verified.

Key words: machine learning, test oracle problem, metamorphic testing, metamorphic relation