Edge computing method for deep neural network

The invention provides an edge computing method for a deep neural network, and belongs to the field of edge computing and deep learning. Compared with an existing model compression scheme and a scheme in which data filling near the segmentation points is not carried out in model segmentation, data filling is carried out on the data near the segmentation points, so that after the feature graphs unloaded to the edge nodes are subjected to corresponding layer calculation, the splicing of data results is kept consistent with that under the condition that the data results are not split, and precision loss is avoided. 本发明提供了一种面向深度神经网络的边缘计算方法，属于边缘计算及深度学习领域。与现有的模型压缩方案以及模型分割中不进行分割点附近数据填充的方案相比，本发明由于对分割点附近的数据进行了数据填充，使得卸载至各边缘节点的特征图在进行相应层计算后，数据结果的拼接与未拆分情况下保持一致，不会产生精度损失。