Abstract: The surface and internal damage of carbon fiber reinforced polymer (CFRP) laminates under 20 J and 40 J impact loads were identified by infrared thermography. Aiming at the inaccuracy of quantitative defect extraction, a multi-scale eight-direction edge detection image segmentation algorithm was proposed by analyzing the spatial characteristics of infrared images. Firstly, the optimal image was selected according to the maximum standard deviation of the sensitive area, and the fuzzy C-means clustering algorithm was used to presegment the defect image to obtain prior information. Then a circular convolution template was constructed to perform a multi-scale eight-direction convolution operation on the infrared image. The OTSU algorithm was introduced to segment the gradient image, combined with morphological operations to obtain the defect edge map, and the connected domain of the target area was analyzed to realize the quantitative extraction of defect features. The results show that the proposed algorithm improves the detection ability of weak edges in the damaged area and ensures the integrity and connectivity of defect edges. Compared with the traditional image segmentation algorithm, the detection accuracy of the defect area, long diameter and short diameter obtained by the proposed algorithm are improved by more than 20.41%, 5.61% and 9.77%, respectively.