Transferable adversarial attack method based on generative adversarial network

The invention provides a migratable adversarial attack method based on a generative adversarial network. According to the method, an auxiliary generator is introduced into a generator to improve generation quality and sample diversity, and features are focused at different scales by using a pyramid attention mechanism. Prior to the generator network, the secondary information is first generated by the secondary generator network. And splicing the generated auxiliary information with the original image, inputting the spliced input image into the generator network, introducing a pyramid attention module in order to further improve the performance of the generator network, allowing the network to pay attention to important areas and features of the image on different scales, and performing processing and conversion through a series of neural network layers so as to improve the performance of the generator network. And a high-resolution vivid image containing disturbance is generated. The image has a higher attac