This paper proposes an efficient real-time semantic segmentation model optimized for embedded platforms to address the challenges of high accuracy, low latency, and energy efficiency in resource-constrained devices. The model features a lightweight Multi-Scale Receptive Field Enhancement (MRFE) Module for improved feature representation, a Pixel-Attention-Guided (PAG) module to guide spatial information, and a Dual Axis Fusion (DAF) module for better feature fusion. A tailored loss function further enhances accuracy. Experimental results show that the model outperforms state-of-the-art models, achieving 49 FPS, 4287 mW energy consumption, and reasonable accuracy (mIoU: 49.13%, Dice: 61.43%) on the LPCVC2023 dataset. Its generalization capability is also validated on the Uavid dataset, achieving 66.42% mIoU and 78.86% Dice.