Cloud-Edge Joint Inference Algorithm for Target Recognition in Cloud-Edge Collaborative Intelligence

Abstract

Cloud-edge co-computing has been widely used for real-time processing of image detection scenarios in IoT. The functional limitations of edge devices and how to effectively determine whether an edge device initiates cloud mode is a new challenge. By proposing a joint inference algorithm. The goal of this algorithm is to automatically determine the trigger conditions for activating cloud mode and improve the effectiveness of cloud-edge collaborative reasoning. The algorithm considers the resource utilization of the edge device, the image recognition accuracy of the edge device, and the network conditions between the edge device and the cloud platform to decide whether to trigger the cloud-edge collaborative inference. Experiments are also conducted on a pathology image recognition project, and the results show that the edge devices show significant improvements in end-to-end reliability and workload reduction.