Flash thermography is a promising technique to perform rapid non-destructive testing of composite materials. However, it is well known that several difficulties are inherently paired with this approach, such as non-uniform heating, measurement noise and lateral heat diffusion effects. Hence, advanced signal-processing techniques are indispensable in order to analyze the recorded dataset. One such processing technique is Gapped Smoothing Algorithm, which predicts a gapped pixel’s value in its sound state from a measurement in the defected state by evaluating only its neighboring pixels. However, the standard Gapped Smoothing Algorithm uses a fixed spatial gap size, which induces issues to detect variable defect sizes in a noisy dataset.

In this paper, a Multi-Scale Gapped Smoothing Algorithm (MSGSA) is introduced as a baseline-free image processing technique and an extension to the standard Gapped Smoothing Algorithm. The MSGSA makes use of the evaluation of a wide range of spatial gap sizes so that defects of highly different dimensions are identified. Moreover, it is shown that a weighted combination of all assessed spatial gap sizes significantly improves the detectability of defects and results in an (almost) zero-reference background. The technique thus effectively suppresses the measurement noise and excitation non-uniformity. The efficiency of the MSGSA technique is evaluated and confirmed through numerical simulation and an experimental procedure of flash thermography on carbon fiber reinforced polymers with various defect sizes.

闪光热成像技术是一种对复合材料进行快速无损检测的有前途的技术。但是，众所周知，这种方法固有地存在一些困难，例如加热不均匀，测量噪声和横向热扩散效应。因此，为了分析记录的数据集，先进的信号处理技术必不可少。一种这样的处理技术是间隙平滑算法，其通过仅评估其相邻像素，根据缺陷状态下的测量来预测其声音状态下的间隙像素值。但是，标准的间隙平滑算法使用固定的空间间隙大小，这会引发问题来检测嘈杂数据集中的可变缺陷大小。

在本文中，引入了多尺度间隙平滑算法（MSGSA）作为无基线图像处理技术，并且是对标准间隙平滑算法的扩展。MSGSA利用对各种空间间隙尺寸的评估，从而可以识别出尺寸差异很大的缺陷。此外，表明所有评估的空间间隙大小的加权组合显着提高了缺陷的可检测性，并导致了（几乎）零参考背景。因此，该技术有效地抑制了测量噪声和激励不均匀性。通过数值模拟和具有不同缺陷尺寸的碳纤维增强聚合物的快速热成像实验程序，评估和证实了MSGSA技术的效率。