Verification code synthesis method based on generative adversarial network

The invention relates to a verification code synthesis method based on a generative adversarial network. The verification code synthesis method comprises the steps of noise input, verification code generator, verification code output, integrated model recognition accuracy calculation, artificial recognition accuracy calculation, verification code addition into a data set, verification code abandoning and verification code data set online application. In the noise input process, random sampling is carried out in a feature representation space and is used for input of a verification code generator; the verification code generator can generate a corresponding verification code according to input noise; the integrated model recognition comprises a plurality of verification code recognizers, and the average accuracy rate of the verification codes generated by the verification code recognizers is calculated; the manual identification is that a plurality of testers identify the generated verification code, and the a