Software architecture of a system for ai-generated image detection based on the xception network

Abstract

Abstract: The paper specifies a software architecture for a system that detects AI-generated images using a configured Xception network. Because the discriminative evidence in this problem consists of low-amplitude generation artifacts, the software architecture is subject to correctness requirements that are not typically encountered in conventional image-classification systems. Three design principles are formulated; based on them, the system is decomposed into four components with stable interfaces. The architectural decisions dictated by the problem are summarized in a table, and the anti-patterns prevented by the proposed architecture are analyzed.