Abstract: ‍ ‍With the development of Beyond 5th Generation （B5G） and 6th Generation （6G） mobile communication networks， the coexistence of cross-modal data， such as visual and tactile data， has emerged as a key challenge. Visual data requires high throughput， whereas tactile data demands 1 ms end-to-end latency and an extremely low packet loss rate （≤10^-5）. The competition for resources between these data types leads to severe allocation fragmentation. To address this issue， this paper proposes a joint optimization scheme based on an overlapping architecture， Non-Orthogonal Multiple Access （NOMA）， and Mobile Edge Computing （MEC）， and designs an efficient resource allocation strategy to enable the coexistence of cross-modal data. First， a cross-modal data coexistence model in the MEC scenario is constructed， decomposing the non-convex optimization problem into two subproblems： subchannel matching and power allocation. For subchannel allocation， an improved Hungarian algorithm is adopted to achieve low-complexity visual-tactile user pairing， overcoming the limitations of traditional one-to-one matching mechanisms in cross-modal scenarios. For power allocation， a closed-form solution is derived to minimize the energy consumption of tactile users. Combined with the Successive Interference Cancellation （SIC） technique of NOMA， the method ensures that tactile signals are decoded first to meet low-latency requirements while minimizing the rate loss of visual data. Additionally， a dynamic iterative algorithm is proposed. By alternately optimizing the subchannel allocation matrix and power vector， it adapts to time-varying channels and bursty services， ensuring convergence to a near-optimal solution in dynamic environments. Simulation results show that， under constraints of visual data rate and tactile data latency， the proposed algorithm reduces the average energy consumption of tactile users by up to 10%， significantly outperforming baseline methods. In a high-user-density scenario （20 users， 10 resource blocks）， the algorithm converges in just two iterations， achieving performance comparable to exhaustive search with a 20% reduction in computational complexity. When the number of resource blocks changes， the algorithm maintains stable energy optimization capability， and tactile energy consumption increases reasonably with growing data volume. Furthermore， improved latency tolerance for visual data can further reduce tactile energy consumption， verifying the adaptability of the algorithm to diverse quality-of-service requirements. Through deep integration of NOMA and MEC and hierarchical optimization strategies， this paper effectively balances energy efficiency， latency， and computational complexity， providing a practical solution for multi-service coexistence in B5G/6G networks. It supports emerging applications such as industrial automation， remote surgery， and immersive visual communication.