Abstract: Semantic communication focuses on transmitting semantic meanings to improve transmission efficiency. A distinct characteristic of this approach is the varying importance of the transmitted information. Unlike the typical bitstream transmission， semantic communication aims to enhance communication efficiency and system performance by conveying the meaning of information instead of bits. In semantic communication， the objective is not to transmit every bit precisely but to reduce redundancy and noise by understanding and interpreting the information. Sparse code multiple access （SCMA） is a representative non-orthogonal multiple access technique based on code-domain superposition. It combines low-density codes with high-dimensional modulation to improve the access and carrying capacity of wireless communication systems via a joint optimization design. Compared with conventional orthogonal access techniques， SCMA can support more user connections and a higher system capacity under the same physical resource conditions through dynamic overloading. Additionally， SCMA is a promising method for reflecting the semantic importance of the physical layer， as it enables unscheduled access and reduces the scheduling overhead of system resources. In this study， we propose an SCMA-based semantic communication framework， where the hierarchical transmission of SCMA is effectively integrated with layered semantic information. The message-passing algorithm （MPA） is the most commonly used detection method in SCMA； however， its computational complexity increases significantly with the codebook size. Hence， we introduce an improved SCMA codebook for semantic mapping， which adjusts the transmission quality of each SCMA information layer based on the importance of semantic information while significantly reducing the codebook size to reduce detection complexity. Furthermore， we propose a low-complexity detection method known as the Proj-IQS MPA algorithm， in addition to its corresponding logarithmic domain algorithm， Proj-IQS Max-Log MPA. Simulation results indicate that the improved codebook outperforms conventional designs in terms of multiscale structural similarity. Specifically， it effectively recovers semantic information at low signal-to-noise ratios while significantly reducing detection complexity.