Sparse code multiple access (SCMA), as a novel representative of non-orthogonal multiple access (NOMA), has garnered significant interest for its capability to support massive device access and improve spectrum utilization in 5G wireless networks. SCMA capitalizes on codebook sparsity to enable multiple users to share limited resources and utilizes the message passing algorithm (MPA) for efficient multi-user detection. Nevertheless, the high computational complexity of MPA, particularly in large-scale access scenarios, poses a significant hurdle for practical implementation. To tackle this problem, this paper proposes a novel detection scheme, referred to as improved serial and sphere decoding MPA (ISSD-MPA), which integrates the concepts of sphere decoding (SD) with a refined serial scheduling strategy. The proposed approach dynamically determines the priority of user node updates according to the maximum message update gain in the factor graph. Moreover, it restricts the decoding search space by pinpointing reliable superimposed constellation points (SCPs) within a threshold radius based on Gaussian distribution. The simulation results demonstrate that the ISSD-MPA algorithm achieves comparable bit error rate (BER) performance to traditional MPA, requiring only two iterations instead of the five iterations typically employed in standard methods. Simultaneously, the computational complexity is reduced by about 70%. Furthermore, the effectiveness of the proposed ISSD-MPA algorithm is validated through simulations and experimental evaluations on the USRP platform, indicating that ISSD-MPA offers a favorable balance between computational complexity and decoding accuracy.