| With the advent of the era of interconnection of all things,a large number of Internet of Things(IoT)devices,namely edge devices,have emerged at the edge of the Internet.At present,the cloud computing center,which has been maintaining a linear growth of computing resources,is gradually unable to handle the explosive growth of data generated by edge devices,edge computing comes into being.With the rapid development of artificial intelligence(AI)technology,cloud computing center pushes part of computing tasks and intelligent services to edge devices,which promotes the process of edge intelligence.Image classification algorithm based on convolutional neural networks(CNN)is one of the most representative algorithms in artificial intelligence technology.At the same time,as a significant part of edge intelligence process,edge devices also need to undertake the tasks of image classification model training and image classification reasoning.Therefore,this paper studies and proposes a coding edge computing scheme for CNN image classification.This scheme combines CNN vectorization,linear coding and distributed computing technology to improve the stability and efficiency of the scheme and protect the security of image data.CNN image classification algorithm is divided into two parts:image classification model training and reasoning.Firstly,aiming at the training process of image classification model,this paper proposes a secure distributed training scheme of image classification model based on heterogeneous edge computing.This scheme extracts a lot of matrix multiplication tasks from the training process of CNN image classification model,and completes the matrix multiplication tasks by using the coding distributed computing strategy on numerous edge devices,which ensures the stability and efficiency of the distributed training process of CNN image classification model,as well as the security of image data.Secondly,aiming at the process of image classification reasoning,this paper proposes a secure distributed image classification reasoning scheme based on heterogeneous edge computing,which is divided into two parts:model distributed deployment and secure distributed image classification reasoning.In the model distributed deployment,the image classification model which is segmented by layers is backed up and deployed on a number of edge devices.In the secure distributed image classification reasoning,the scheme adopts a variety of distributed reasoning schemes to ensure the efficiency of distributed image classification reasoning and the security of image data.Finally,this paper proposes a coding edge computing scheme for CNN image classification based on the above two schemes.This scheme adopts NumPy library to speed up matrix multiplication,and uses Python programming language to realize an efficient edge distributed computing system.The results of simulation experiments show that the proposed scheme has high stability and efficiency,which can reduce the time of CNN image classification model training and reasoning on heterogeneous edge devices and can protect the security of image data. |
