Scale-Aware and Boundary-Enhanced Semantic Segmentation

Wei Zhou

Abstract: Semantic segmentation is widely used in autonomous driving and medical imaging, but accurately segmenting objects with different scales and preserving object boundaries remain challenging. This paper proposes a semantic segmentation network that combines a Multi-Scale Feature Fusion (MSFF) module with a Boundary Enhancement Module (BEM). The MSFF module captures contextual information using multiple receptive fields to improve multi-scale object recognition, while the BEM exploits shallow edge features and spatial attention to enhance boundary reconstruction. Experiments on the PASCAL VOC 2012 and Cityscapes datasets demonstrate that the proposed method achieves competitive segmentation performance and improves boundary accuracy compared with representative baseline methods.

Keywords: Semantic Segmentation, Multi-Scale Feature Fusion, Boundary Enhancement, Spatial Attention, Deep Learning, Computer Vision