ABSTRACT

This study introduces and evaluates a new approach to reconstruct image sequences acquired during non-destructive testing by pulsed thermography. The proposed method consists of applying two linear support vector regressions to model the evolution of the data from both a spatial and temporal point of view. Each regression vectors will map the data with the number of pixels and the number of frames using convex optimisation. Then the regression vectors are used to predict a more robust representation of the data, thus reconstructing the sequence. The proposed method has been applied to data related to a reference sample of carbon reinforced fibre with known defects. This approach was evaluated on a sequence with severe non-uniform heating and was compared with state-of-the-art methods. Despite being sensitive to non-uniform heating, the proposed method provided a higher CNR score on smaller defects, compared with state-of-the-art methods. For the shallowest defects it shows better performance in term of contrast reconstruction compared to partial least-squares thermography (PLST). It also outperforms principal component thermography (PCT), and thermographic signal reconstruction-PCT (TSR-PCT) for defects located at a depth of 0.6 mm and 0.8 mm below the surface.

摘要

本研究介绍并评估了一种重建在脉冲热成像无损检测期间获取的图像序列的新方法。所提出的方法包括应用两个线性支持向量回归来从空间和时间的角度对数据的演变进行建模。每个回归向量将使用凸优化将数据与像素数和帧数进行映射。然后使用回归向量来预测更稳健的数据表示，从而重建序列。所提出的方法已应用于与具有已知缺陷的碳增强纤维参考样品相关的数据。该方法在具有严重不均匀加热的序列上进行了评估，并与最先进的方法进行了比较。尽管对不均匀加热很敏感，与最先进的方法相比，所提出的方法在较小的缺陷上提供了更高的 CNR 分数。对于最浅的缺陷，与部分最小二乘热成像 (PLST) 相比，它在对比度重建方面显示出更好的性能。对于位于表面以下 0.6 毫米和 0.8 毫米深度的缺陷，它的性能也优于主成分热成像 (PCT) 和热成像信号重建-PCT (TSR-PCT)。