针对脉冲涡流热成像检测技术中原始红外图像信噪比较低、温度对比性较差以及存在邻近效应和不均匀加热的问题, 将奇异值分解法应用到热图序列的处理中以增强重构图像中的缺陷特征。介绍了奇异值分解法的原理, 用奇异值分解法对实验中采集到的热图序列进行处理, 以信噪比为指标对图像的处理效果进行评定。结果表明奇异值分解法能够抽取红外热图序列反映缺陷信息特征, 可消除邻近效应和不均匀加热的影响, 提高图像的信噪比。将奇异值分解法与主成分分析法比较, 发现前者重构的图像质量高于后者, 是处理红外热图序列的又一有效方法。

The raw thermal images used in pulsed eddy current thermographic nondestructive testing was characterized with low signal-to-noise ratio(SNR) and temperature contrast. Additionally, there are still uneven heating and proximity effect problems. To avoid the performance degradation caused by these issues, singular value decomposition(SVD) technique was applied in infrared image sequence processing to enhance characteristics of the reconstructed image defects. The SVD was introduced and was used to process infrared images. The image processing effect was evaluated by SNR. The result shows that SVD can extract the algebra characteristics to reflect defect information. It can remove the uneven heating and proximity effects, and increase the SNR of images. Images reconstructed by SVD were compared with those by principal component analysis (PCA). It demonstrated that SVD outperforms PCA in terms of reconstructing images, thus SVD is an effective method to process infrared image sequence.