Fig 3  Comparison between original and optimized channel shuffle mode.
Experiments and results: The network we proposed is trained and tested using the PyTorch deep learning framework. The test dataset we used is CIFAR100. The simplified ShuffleNetV2 network proposed in this paper is compared with traditional networks, and the comparison results are shown in the table Ⅰ. In the case of small input feature map size, the network achieves a recognition accuracy improvement of 1.33% and 1.09% compared to the original network. At the same time, the number of parameters is reduced from 1.4M to 1.22M, effectively reducing the storage resources and data access on the hardware.