Objective Aiming at the problem that the high data traffic in Data Center Network DCN leads to the low recognition accuracy of elephant flow and rat flow a two-stage elephant flow recognition algorithm based on Software Defined Networking SDN was proposed. Methods Combining SDN and DCN the dynamic threshold optimization algorithm of Gaussian distribution was adopted in the first stage to calculate the false detection rate and missing rate of elephant flows by setting the packet threshold and the optimal threshold was continuously optimized to identify the suspicious elephant flows. In the second stage based on the accurate elephant flow obtained from the suspicious elephant flow in the previous stage according to the transmission rate and duration of the flow the threshold constraint traffic detection mechanism and Count counter were proposed to improve the constraint of the lower limit of the elephant flow identification threshold and the data volume and duration of the elephant flow in the network were calculated as the threshold value within the cycle so as to improve the identification accuracy of elephant flows. Results The experimental results showed that compared with the existing algorithms the average number of bytes of the suspicious elephant flows in the first stage was 11. 3% more than that of the network flows. Under different thresholds the accuracy of the algorithm was increased by 1. 7% and the average detection time under different network traffic was reduced to less than 6 ms. Conclusion The two-stage elephant flows recognition algorithm in the software-defined network has strong elephant flow recognition ability in the first stage. At the same time the accuracy of the algorithm is improved the average detection time of the elephant flows is reduced and the network quality is improved providing relevant conditions for further study of the network traffic scheduling strategy.