<p>Edge intelligence represents an emerging computing paradigm that processes data in real time and performs artificial intelligence tasks at the network edge. Using containerized serverless computing on edge devices can further achieve more real-time and efficient computing. However, the limited and heterogeneous computing resources in the edge, coupled with the geographical dispersion of devices and data, have seriously hindered the advancement of edge intelligence. Consequently, how to fully utilize the computing resources at the network edge while ensuring low delay of containers is a major challenge currently faced. To address this challenge, this paper proposes a resource clustering-based container scheduling (RCCS) algorithm. The RCCS achieves efficient heterogeneous resource management through resource clustering and models the container scheduling problem as a multi-criteria decision-making problem. The problem of solving the weight coefficients between the criteria is transformed into a mixed-integer linear programming problem. The optimal weight vector is then obtained through the CBC solver. Subsequently, the scores of each edge node can be calculated, and the container request is scheduled to the node with the highest score for execution. The results of a large number of simulation experiments based on real-world data demonstrate RCCS achieves an effective trade-off between reducing delay and improving edge resource utilization. Compared with existing container scheduling algorithms, the RCCS improves the average GPU utilization of edge clusters by 23.54% and reduces the average completion time of containers by 18.35–31.44%.</p>