 Pulse thermography is a powerful tool used in the field of non-indistructive evaluation, NDE. However, analyzing the large amount of data generated from this technique requires expertise and can be time-consuming. Deep learning has made significant advances in image segmentation, which has led to the development of deep learning models for NDE applications such as pulse thermography. To evaluate the performance of these models, a new dataset was created with 1,000 images and 2,500 frames of data. Additionally, the effects of different frame numbers and data transformations were studied. The results showed that pre-processing techniques can significantly reduce the size of the data without affecting model performance. This suggests that deep learning methods can be used to make NDE data analysis faster and more accessible to a wider range of practitioners. This article was authored by Ziang Wei, Ahmad Osman, Bernvelesque, and others.