Empowered edge intelligent aquaculture with lightweight Kubernetes and GPU-embedded

Edge computing is a new paradigm for processing data at the edge of networks. There are a variety of edge computing scenarios, depending on the situation. In this paper, we investigate an architecture with heterogenous devices for intelligence aquaculture. The system will collect water sensor data and run real-time video-based fish detection with a Deep Learning algorithm and Deepstream. Each system was monitored to ensure all the architecture design was running properly. Kubernetes Lightweight Kubernetes (K3s) was used to manage all applications deployed in Docker containers. In addition, to synchronizing the heterogenous devices and visualizing the node’s resource system, the Rancher Kubernetes Engine is used to coordinate its resources. The container-based architecture with embedded GPU can work properly for fish detection and the POD schedule scheme that we propose show an improvement in container AI performance and increased around 20%. The architecture of this system can be referenced as a model for edge computing ecosystems.