Abstract: In response to the urgent demand for high reliability and high precision in non-destructive testing of welds in the aerospace manufacturing field, and aiming at the problem of low accuracy in defect recognition of low-contrast radiographic images caused by the lack of domain knowledge in existing weld defect recognition methods, an innovative weld defect recognition method based on feature knowledge in the aerospace field is proposed. By analyzing the types and characteristics of weld defects, a feature knowledge system in the aerospace field with clear physical meaning and explainable is constructed to bridge the semantic gap between process knowledge features and image features, and improve the accuracy and credibility of recognition algorithms. The experimental results show that for the aerospace weld X-ray dataset, the average recognition accuracy of this method reaches 93%, which is 14% and 7% higher than that of the traditional models (ViT-base/16 and Swin-S), indicating that the knowledge in the aerospace field is deeply integrated with the deep learning model in the form of structured features. It can effectively enhance the credibility and accuracy of weld defect identification in the aerospace field, providing a feasible technical path for improving the quality of aerospace products.