Abstract: In order to improve the efficiency of software automated testing, this paper proposes a set of automatic test architecture based on artificial intelligence recognition model, model resource pool and scene fusion adaptation method. Firstly, the artificial intelligent recognition model is used to realize the identification and localization of the operation object. Secondly, the model resource pool is used to further improve the recognition accuracy and generalization ability. Finally, the scene fusion adaptation method is used to support different levels of automation execution requirements. This architecture realizes the automatic testing capability of the software system covering multiple types of terminals without the need for script recording, and providing support for improving quality and efficiency in various related application scenarios.