A Semantic Segmentation Method for Road Sensing Images Based on an Improved PIDNet Model
Guangxing Tan, Yangying Jin · Electronics · 2025
Semantic segmentation, as a critical technology in intelligent transportation and autonomous driving, plays a significant role in accurately parsing scene information and enhancing environmental perception capabilities. However, the complexity of road environments poses challenges to the robustness and real-time performance of existing algorithms. Although PIDNet achieves a certain balance between performance and efficiency, it still falls short in fine-grained object segmentation and multi-scale feature fusion. To address these issues, this paper proposes an improved algorithm based on PIDNet. The proposed method includes the following: (1) introducing a multi-branch high-resolution feature extraction module to reduce information loss; (2) adopting a dense atrous spatial pyramid pooling module to enhance multi-scale feature fusion capabilities; and (3) incorporating cross-attention mechanisms into the Bag module to optimize feature interaction. Experimental results on the CityScapes dataset show that the improved algorithm increases the mean intersection-over-union (MIoU) from 78.6% to 81.1%, demonstrating higher segmentation accuracy and robustness in complex scenarios while maintaining real-time performance, thereby validating the effectiveness of the approach.