The debonding of non-metallic materials can affect their performance, and infrared non- destructive testing can identify adhesive defects effectively. The boundary feature of adhesive defects is first investigated in this study based on infrared non-destructive testing; subsequently, a quantitative analysis method for identifying the boundary position of adhesive defects using the extreme value of a temperature gradient is obtained. Next, the Canny edge detection algorithm is used to identify the defect boundary of an adhesive defect model by numerical simulations and to identify experimental data simultaneously. For problems such as blurred boundaries and noise for recognition results, an improved algorithm for filtering out all “suspected boundaries” to preserve the “weak boundary” is proposed. The results show that the improved Canny algorithm can improve the integrity and accuracy of identifying adhesive defects from infrared non-destructive testing.