Abstract:The welding quality of aircraft truss is an important guarantee of its working strength, so the effective detection and identification of truss weld defects is the focus of current aviation manufacturing industry. Aiming at the problems of complex calculation and low recognition accuracy existing in traditional target recognition methods, in order to detect the internal defects of weld quickly and effectively, a weld defect recognition method based on improved convolutional neural network (CNN) is proposed. Firstly, the threshold value of weld image is divided to make the feature information easier to extract. Then, an improved adaptive pooling method is designed, and a new model structure of weld image defect recognition is proposed, and the corresponding model parameters and calculation method are formulated. Finally, the recognition model is used to train and test the weld image. The results show that the network model can effectively realize the recognition and classification of weld defects, and the average correct recognition rate is 98.25%, which shows that the proposed method has the advantages of fast recognition speed, high accuracy and good robustness, and provides theoretical reference for the process of weld defect recognition.