Abnormal Object Detection System in Cloud Server Centers Based on Transformer

In order to solve the problem of foreign object detection in the cloud server center, this paper proposes an abnormal object detection system based on a transformer network, aiming to identify and warn abnormal objects in the computer room environment. Firstly, this paper constructs an anomaly object dataset (NEU AID) to contain information about the computer room environment, and it improves the diversity and generalization ability through data enhancement technology. Secondly, an improved Dual Flow Transformer (DFT-NET) is proposed, which can extract both global and local features to optimize the detection effect. This paper introduces transfer learning and uses the COCO data set to pre-train the network. Finally, through practical application verification, the proposed DFT-NET network model shows the high accuracy identification ability of unknown abnormal items when the inspection robot performs the abnormal items detection task in the computer room, proving the model's effectiveness and reliability in practical application scenarios.