Weld X-ray image defect recognition based on deep learning

Abstract

This paper discusses the key link in weld quality assessment —— weld defect feature extraction and defect identification based on BP neural network model. Firstly, the main defect classification of welds and the technical method of feature extraction are introduced in detail, and six feature parameters are proposed and experimental simulation is performed. In terms of image preprocessing, the noise reduction technology based on domain averaging method and segment linear transformation are adopted to enhance the contrast and details of the image, and the weld path area is accurately extracted and segmented by line grayscale curve method and iterative threshold segmentation method based on subtraction technology. Then, the principle and key parameters of BP neural network model are introduced in detail, the optimal model parameters are selected through tuning and training, and the performance of the model is evaluated. Finally, the classification results of the test set are analyzed, demonstrating the performance of the model in identifying cracks and stomata. The experimental results showed that for the test set, the crack defect recognition rate was 100% and 73.3%. A quality and defect identification program, which provides important theoretical support and practical guidance in the field of welding engineering.