Abstract: The ultrasonic excited thermography nondestructive testing was utilized in detecting an impact damage of a composite material. Considering the thermal images characteristics of low contrast und edge fuzziness, a LBF model based on a non-Gaussian kernel function was used to detect the defect edge and a quantitative calculation was performed. The experiment result showed that the LBF model could effectively identify the object edges of low contrast thermal images, comparing to other common edge detection methods, which laid a good foundation for the quantitative assessment of composite defects. What is more, the detection accuracy of defect length and width showed that ultrasonic excited thermography nondestructive testing was an effective detect method for the defect detection of composite material, which provided powerful technical support for research and development of composite materials.