For the characteristics of marginal environment such as battlefields which environment is very difficult to measure, environmental perception has difficulty, resources are scarce, communication conditions are inferior and energy issues are highlighting. Depending on the issue that current architecture cannot process data in real time the edge calculation technique is proposed and aims to address the shortcomings of cloud computing in processing data that cannot be remedied. According to the EdgeX open source framework, an edge computing platform is designed in this paper. And realizes the three-layer network structure of “terminal edge-cloud” or “terminal-intelligent gateway-edge cloud-cloud”. According to the design idea of loose coupling, the edge computing platform realizes the loose coupling between microservices and the separation of management and service. A resource scheduling strategy based on depth Q-learning is used to realize adaptive scheduling, which can effectively deal with the fluctuation of load demand and maximize the utilization of resources.