Method for augmenting image based on generative adversarial cascade network

Abstract

The invention discloses a method for augmenting an image based on a generative adversarial cascade network. The method comprises the steps of delineating a region of interest from an original image Iori and cutting the region of interest to obtain a cut image Icut; preprocessing the Icut to obtain an augmented data set Scut; training the I-level generative adversarial network by using the data setScut; loading the trained I-level generator, inputting random noise to infer an image, and performing up-sampling processing on the generated image to make a new data set S1; using the data set S1 and the data set Icut as a training data set of the II-level generative adversarial network, and performing training of the II-level generative adversarial network; and loading the trained II-level generator, inputting the data set S1 into the II-level generator, and reasoning a required augmented image Ides. According to the method, when the image is augmented, the problems of small difference andlow resolution of the generated image in the I-level generative adversarial network are solved, and the generalization performance of the network is improved while the image is augmented.

Description

technical field [0001] The invention relates to the field of ultrasonic image analysis, in particular to a method for augmenting an image based on generating an adversarial cascade network. Background technique [0002] In the image research of deep learning, it usually relies on large-scale data sets to avoid the occurrence of over-fitting problems. When the amount of image data is seriously insufficient, traditional image augmentation methods are usually used for image augmentation, such as multiple cropping, adding Gaussian noise, gray balance, etc. [0003] These traditional image augmentation methods not only expand the existing data sets, but also bring the risk of over-fitting to the training of the network. The reason is that the images obtained by these traditional image augmentation methods are highly correlated with the original images, and the images generated by the single-stage generative adversarial network also have certain similarities and low resolution. T...

AI Technical Summary

Problems solved by technology

The reason is that the images obtained by these traditional image augmentation methods are highly correlated with the original images, and the images generated by the single-stage generative adversarial network also have certain similarities and low resolution. These methods cannot significantly improve the data set. sample diversity

As the amount of augmented data increases, there are more and more similar data items in the data set, which eventually leads to over-fitting of the network and poor generalization performance.

[0004] In the field of deep learning, there are often situations where the amount of image data is insufficient or the types of images are not rich enough. Using a good image augmentation method can often achieve twice the result with half the effort or even a decisive role; but at the same time, a single image augmentation method It may also lead to over-fitting of the network, resulting in poor generalization performance of the network; in addition, the images generated by the single-stage generative confrontation network have problems such as high similarity between images and low resolution

Images