High-mobility image confrontation sample generation method based on generative model
The invention relates to a high-mobility image confrontation sample generation method based on a generative model. The method comprises the steps that sample images of a black box target model training set are collected and preprocessed to obtain training samples; constructing and training a white b...
Saved in:
Main Authors | , , , |
---|---|
Format | Patent |
Language | Chinese English |
Published |
27.09.2022
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The invention relates to a high-mobility image confrontation sample generation method based on a generative model. The method comprises the steps that sample images of a black box target model training set are collected and preprocessed to obtain training samples; constructing and training a white box substitute model, and constructing a generative network model comprising a disturbance reutilization module PRM and a feature enhancement module FEM; using the trained white box substitute model to train and generate a network model; and inputting the target image into the trained generative network model, generating an adversarial sample of the target image by the trained generative network model, and inputting the adversarial sample of the target image into the black box target model to realize the black box attack based on the mobility of the adversarial sample. According to the method, the middle layer features of the white-box substitute model are effectively utilized, overfitting of the countermeasure samp |
---|---|
Bibliography: | Application Number: CN202210663143 |