Abstract A novel algorithm for pixel-based thermal sequences processing based on R2 (coefficient of determination, COD) fractile threshold of non-linear fitting is proposed. Original data are obtained from active laser infrared thermography inspection to an aviation CFRP laminate with artificial defects. By fitting temperature sequence of pixels in a thermogram using the 1st-order and 2nd-order Fourier function, R2 is calculated and attached to each pixel. It is found that R2 of defective pixels is greater than that of non-defective pixels. The threshold for filtering background pixels is determined by histogram analysis of R2 set in a thermogram. Then fractile thresholds of R2 set separate pixels from defect and non-defect region. The temperature value of non-defect pixels is reduced to contrast with defect pixels. The edge of defects is sharpened, and detectable defects increase. Combining the fractile thresholds of 1st-order and 2nd-order Fourier fitting, the final processed thermograms exhibit improved detection performance evaluated by simplified probability of detection (POD) analysis and signal-to-noise ratio (SNR) calculation. In this work, the tested data show huge promotion of defect detection rate. Defect with an aspect ratio of 3.0 is detected. The combine-processed thermogram possess the maximum SNR of defective area.

中文翻译：

---

基于R2分形阈值的主动红外热成像非线性拟合的像素热序列处理算法

摘要 提出了一种基于R2（coefficient of decision，COD）非线性拟合分位数阈值的基于像素的热序列处理算法。原始数据是通过主动激光红外热成像检查对具有人工缺陷的航空 CFRP 层压板进行检测获得的。通过使用一阶和二阶傅立叶函数拟合热图中像素的温度序列，计算 R2 并将其附加到每个像素。发现缺陷像素的R2大于非缺陷像素的R2。过滤背景像素的阈值由热谱图中设置的 R2 的直方图分析确定。然后 R2 的分形阈值将像素与缺陷和非缺陷区域分开。降低非缺陷像素的温度值以与缺陷像素形成对比。缺陷边缘锐化，可检测缺陷增加。结合一阶和二阶傅立叶拟合的分位数阈值，最终处理的热谱图展示了通过简化检测概率 (POD) 分析和信噪比 (SNR) 计算评估的改进检测性能。在这项工作中，测试数据显示了缺陷检测率的巨大提升。检测到纵横比为 3.0 的缺陷。组合处理的热谱图具有缺陷区域的最大信噪比。测试数据显示缺陷检出率有巨大提升。检测到纵横比为 3.0 的缺陷。组合处理的热谱图具有缺陷区域的最大信噪比。测试数据显示缺陷检出率有巨大提升。检测到纵横比为 3.0 的缺陷。组合处理的热谱图具有缺陷区域的最大信噪比。