Ultrasonic Image Processing Technology Based On Deep Dictionary Learning

Deep dictionary learning combines the advantages of deep learning and dictionary learning,builds a deep structure through multi-level dictionary learning,adopts a greedy learning model,and ensures the convergence of each layer of dictionary through layer-by-layer training.The deep dictionary learning model is used to learn the image data,and the image data can be accurately represented according to the obtained dictionary.Applying the deep dictionary learning model to the acquired ultrasonic image processing,combined with ultrasonic microscopy imaging technology can achieve a more accurate representation of the internal feature information of the flip chip,which is important for the location and analysis of defects in the ultrasonic inspection image significance.On the basis of deep dictionary learning,the thesis has carried out research on the internal imaging of flip-chip chips.The main contents include:(1)In order to dig out the deep-level feature information of image data,the existing deep learning model and dictionary learning algorithm are studied,and the deep dictionary learning model is designed to represent the image data,and a dictionary containing image data features is obtained.The deep dictionary learning model is used to realize the reconstruction of the image on the natural image data set.The results of the stacked autoencoder model,the single-layer dictionary and the KSVD reconstruction image are compared.The experimental results verify that the dictionary obtained by the deep dictionary learning is not The reconstruction of image data has good performance.(2)A SMP-AMI imaging model based on deep dictionary learning is designed.The deep dictionary learning model is used to learn and process the two-dimensional B-scan ultrasound image to obtain the dictionary,and the feature library is combined with the sparse representation-based AMI imaging technology to image the interface of the flip chip.Through the comparison of experimental results,the results of the deep dictionary learning model combined with SMP-AMI imaging are more accurate than the single-layer dictionary learning and the imaging results of the KSVD dictionary combined with SMP-AMI imaging technology to display the defect location.