The implementation and application verification of an intelligent recognition system oriented toward edge computing

Abstract

With the development of the Internet of Things and artificial intelligence, intelligent recognition systems have gradually appeared, but there are still serious problems of resource limitations and changes in the operating environment at the edge. Research on intelligent recognition at the edge has been carried out, and in this paper, a distributed computing architecture with full-stack cloud cooperation is put forward. The two are far apart, but they are not the same and have different purposes. To address the problem of model deployment, structured pruning, knowledge distillation and quantization-aware training are employed to reduce the size of the model, and then dynamic computation graph optimisation is used for online model deployment. Based on the attributes of recognition tasks, a model to predict resource demand and a dynamic-priority scheduling method with latency constraints have been developed, and an adaptive update method for data drift has also been added to ensure the normal operation of the system in the face of all sorts of changes in resource conditions. All-round technical solutions have been put forward in the above studies to solve the problems of recognition accuracy, resource utilisation and real-time operation effectively in the construction of edge intelligent recognition systems.