Self-adaptive normalization-based unsupervised attention generation network structure and method

Abstract

The invention discloses a self-adaptive normalization-based unsupervised attention generation network structure and method. A Cycle-GAN is used as an infrastructure, the network structure comprises a generator and a discriminator, the generator mainly comprises a high and low frequency convolution layer, a self-attention layer and a self-adaptive normalization layer, and the high and low frequency convolution layer is used for extracting first-order information from an input image; the self-attention layer continues to process the extracted first-order information to form second-order image representation with an attention mechanism; the self-adaptive normalization layer is used for cascading the second-order image representation information with the attention mechanism to form final image representation; the discriminator classifies the representation use of the real image and the generated image to discriminate the generated image and the real image so as to assist the generator in better generation. The image features are extracted and recovered, and the extracted features are converted into corresponding images.

Description

technical field [0001] The invention relates to the technical fields of medical image processing, active health, convolutional neural network, and attention mechanism adaptive normalization, in particular to an unsupervised attention generation network structure and method based on adaptive normalization. Background technique [0002] In recent years, computer vision has developed strongly in the field of medical image analysis, and is widely used in a series of application fields such as lesion detection and classification, semantic segmentation, etc. The development of deep learning algorithms, especially the application of convolutional neural network (CNN) and GAN network, the enhancement of image registration and image enhancement related techniques have led to the vigorous development of methods for generating and transforming image data. From another perspective, radiology scans are an important tool in modern medicine. They support diagnosis, disease tracking, and p...

AI Technical Summary

Problems solved by technology

Furthermore, the use of MRI-based radiotherapy is limited in the context of the increasing use of metal implants such as cardiac pacemakers and artificial joints in an aging society

Images