With the rapid development of edge computing and 5G, Kubernetes has become the de facto standard for container orchestration, yet its default scheduler underutilizes resources and causes task imbalance in heterogeneous environments. To address this, we propose SEACO-Scheduler, an improved ACO-based algorithm that integrates node resources, task requirements, and network conditions. By refining pheromone updates and introducing stochastic pheromone dropout, SEACO-Scheduler avoids local optima and balances allocation. In three-node simulations comparing default, round-robin, Q-learning, deep learning, classic ACO, and SEACO-Scheduler, results show that the default, Q-learning, and deep learning methods suffer severe bias, round-robin balances tasks but neglects node capacity, and classic ACO improves utilization yet risks convergence traps. SEACO-Scheduler optimally exploits resource-rich nodes, maintains stable fitness, and achieves superior utilization and system efficiency.