Multi-feature information fusion in ultrasonic phased array for enhanced weld defect identification

Ultrasonic phased array technology is widely used for non-destructive weld defect detection due to its excellent performance. However, Manual identification of microscopic images is rather difficult and cannot be qualitatively identified. Therefore, this study proposes a welding image recognition method based on multi-feature information fusion. Initially, the paper introduces a method for processing the image dataset, including manual cropping, color substitution, filter denoising and threshold segmentation. Subsequently, deep learning is used to automatically extract features from preprocessed defective ultrasonic images and assess how different convolutional neural networks affect feature extraction in multi-view images. MobileNet-V2 is selected as the convolutional neural network for image feature extraction. Ultimately, an intelligent defect image recognition model is constructed based on a multi-view image feature fusion algorithm. Experimental results demonstrate that the A+B and A+B+S view fusion schemes significantly improve model performance. Additionally, an improved attention mechanism is introduced in the feature fusion stage of multi-perspective images to balance information interaction and independence between images from different perspectives. The experimental results show that the performance of the improved model was further enhanced with an accuracy rate as high as 88% for defect classification using the improved A+B+S view feature fusion model.