Parametric Model of Pipe Defect Description for Generation of Training Set for Machine Learning in Data-Poor Conditions

The article addresses the problem of the lack of real data in training and testing of machine learning algorithms. The issue is presented below through the case of identifying the defect on the pipe inner surface. In this paper, the authors present approaches to the formation of a training set using synthetic images obtained from various sources. The article considers in detail the method of image generation based on the recommended parametric representation of the defect on the inner surface of a pipe. For the defect description, the following parameters were selected: area coefficient, HSV color model, texture, shape and boundary of the defect. The determination of each of the selected parameters is described in the paper. The experimental results on the synthetic image generation based on parametric representation of the defect using the developed software in the Matlab environment are noted. The article considers the method of detecting defects on the inner surface of the pipe using the presented defect parametric description. Based on the developed model, there was formed a sample of tube images for the neural network training and testing.

[1]  В. И. Шахуро,et al.  Синтез обучающих выборок для классификации дорожных знаков с помощью нейросетей , 2018 .

[2]  Tal Hassner,et al.  Regressing Robust and Discriminative 3D Morphable Models with a Very Deep Neural Network , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[3]  Feng Liu,et al.  Disentangling Features in 3D Face Shapes for Joint Face Reconstruction and Recognition , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[4]  Kate Saenko,et al.  Learning Deep Object Detectors from 3D Models , 2014, 2015 IEEE International Conference on Computer Vision (ICCV).

[5]  Thomas Brox,et al.  U-Net: Convolutional Networks for Biomedical Image Segmentation , 2015, MICCAI.

[6]  Ming-Ting Sun,et al.  Semantic Instance Annotation of Street Scenes by 3D to 2D Label Transfer , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[7]  Louahdi Khoudour,et al.  Learning to Recognize 3D Human Action from A New Skeleton-based Representation Using Deep Convolutional Neural Networks , 2018, ArXiv.

[8]  Pieter Abbeel,et al.  InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets , 2016, NIPS.

[9]  Nizhny Novgorod,et al.  IMAGE SEGMENTATION METHODS IN PROBLEMS OF SURFACE DEFE CT DETECTION , 2012 .

[10]  Yoshua Bengio,et al.  Generative Adversarial Networks , 2014, ArXiv.

[11]  Louahdi Khoudour,et al.  Learning to recognise 3D human action from a new skeleton-based representation using deep convolutional neural networks , 2018, IET Comput. Vis..

[12]  J. Gielis.  A generic geometric transformation that unifies a wide range of natural and abstract shapes. , 2003, American journal of botany.