Abstract: With the accelerating trend toward intelligent and unmanned coal mines, underground inspection robots are playing an increasingly vital role in environmental monitoring, equipment inspection, and safety assurance. Inspection tasks in underground coal mines exhibit significant variations in both data volume and computational complexity, resulting in a typical dual heterogeneity. Traditional MEC (Multi-access Edge Computing) offloading methods often treat tasks as black boxes, making it difficult to accurately characterize actual computational costs, thereby limiting the effectiveness of offloading decisions.To address this, this paper first constructs an interpretable computational load model based on the task's time complexity and data volume, and derives a critical threshold to distinguish between local execution and offloading. This enables precise selection of offloading candidate tasks. Building on this, a two-layer optimization framework is proposed: the upper layer makes preliminary offloading decisions under the assumption of sufficient resources, while the lower layer performs resource-aware scheduling using a Weighted Satisfaction Resource Management Algorithm (WSRMA) under actual resource constraints. WSRMA integrates weighted satisfaction, resource penalties, and neighborhood search to jointly optimize latency, energy consumption, and load balancing.Experimental results demonstrate that the proposed method improves system-wide weighted satisfaction by 9.49%, while reducing average latency and energy consumption by 7.63% and 3.94%, respectively, across various dual-heterogeneous task scenarios. The method significantly outperforms existing strategies and offers low computational complexity, making it well-suited for online scheduling applications.