Design And Research Of Image Recognition System Based On Edge Computing

In recent years,depth learning has been at the forefront of visual computing.Deep learning improves the computational efficiency of some complex tasks,such as face recognition,object recognition,motion recongnition.With the development of mobile Internet technology and 5G technology.The development of deep learning networks makes it more attractive to mobile devices and wearable devices.However,due to the weak computing ability of wearable devices,it takes more time for deep learning network models to make inferences in wearable devices with limited resources,and deep learning network models will also take up more storage space of wearable devices.Due to high latency or lack of connectivity,offloading the computing of wearable devices to the cloud is not a good solution.How to solve the above problems effectively is still the current research hotspot.In order to solve the these problems,we start from the perspective of actual application deployment and combines depth learning and edge computing technology in this paper,proposes a cooperative computing strategy between mobile and edge,called big/little model cooperative strategy.The strategy reduces the computing pressure of mobile devices and ensuring the accuracy of system identification by offloading the computing of the depth learning model at the mobile to the edge.In addition,in order to solve the problem of unstable network communication in some special application environments,such as earthquake search and rescue,we proposes a cooperative strategy of edge-model in complex network environment in this paper.The strategy adaptively schedules the system according to different network states,so the system can still work independently in case of communication failure,also improving the robustness of the system.Finally,in order to solve the problem of acquiring images due to camera angle,limited field of view,we combines image feature detection and some related technologies of edge computing,proposes an image fusion strategy based on edge computing in this paper.In this paper,the big/little model cooperative strategy and the edge-mobile model cooperative strategy in complex network environment are evaluated by using the cifar and Minst common data sets.The experimental results show that the big/little model cooperative strategy reduces the computation of mobile devices by89.80% and 63.7% respectively on the cifar and Minst data sets without improving the depth learning algorithm.In the non-public data set,the calculation of mobile devices is reduced by 29.34%.The cooperative strategy of edge-model in complex network environment can keep the system running in the environment of network failure,reducing 62.6% and 96.63% transmission delay of mobile devices and edge respectively in cifar and Minst data sets.