This paper investigates the performance of two advanced deep-learning models, YOLOv8 and YOLOv10, in the context of defect detection in composite materials using active thermography. The study aims to address the critical need for accurate and efficient non-destructive testing methods in industrial applications. This paper presents extensive experiments to compare the models in terms of detection accuracy and processing speed. The results reveal that YOLOv8 achieves superior detection accuracy, making it highly effective for applications that require precise defect identification. In contrast, YOLOv10 gives significantly faster processing times, positioning it as the optimal choice for real-time detection scenarios where speed is crucial. The findings of this study provide valuable insights for selecting appropriate models based on the specific demands of defect detection tasks, contributing to the ongoing advancement of deep learning applications in non-destructive testing.