Unsupervised cross-domain target detection and recognition system based on deep

Abstract

The application of target detection and recognition technology in the industry has promoted the further development of computer vision technology. However, for target detection in complex environments, the influence of shooting angle, character pose and lighting environment in different scenes leads to unsatisfactory target detection performance, inconsistent distribution among different scene data domains, and difficulty in migrating existing models to be applied in new target domains. Therefore, for the complex environment, how to effectively solve the problem of model reuse under unsupervised conditions, this paper proposes an unsupervised cross-domain target detection and recognition system based on deep migration learning, which provides a simple operation interface, and through simple and convenient operations can quickly deploy the target detection environment in complex environments, realizing the detection, localization, early warning alarm and logging of vehicles, people, etc. in complex environments storage, etc. Finally, the reliability of the system in practical applications is verified through target detection and character recognition experiments.