2026· Computers, Materials & Continua· Vol 89, pp. 1-10· 0 citations· 39 references
TL;DR
BFANet is proposed, a boundary-aware lightweight semantic segmentation framework based on DeepLabV3 + with a MobileNetV2 backbone that integrates parameter-free SimAM feature refinement, low-level-guided Dynamic Feature Alignment, and progressive decoder fusion to enhance discriminative feature responses, reduce cross-level feature inconsistency, and recover fine boundary structures.
Abstract
: Lightweight semantic segmentation remains challenging because compact backbones often weaken feature discriminability and lose fine-grained boundary details. In DeepLabV3 + -style encoder-decoder architectures, the direct fusion of high-level semantic features and low-level spatial features may introduce semantic-spatial misalignment, resulting in blurred object contours and fragmented predictions. To address these issues, this paper proposes BFANet, a boundary-aware lightweight semantic segmentation framework based on DeepLabV3 + with a MobileNetV2 backbone. BFANet integrates parameter-free SimAM feature refinement, low-level-guided Dynamic Feature Alignment, and progressive decoder fusion to enhance discriminative feature responses, reduce cross-level feature inconsistency, and recover fine boundary structures. Experiments on a curated PASCAL VOC 2012 subset demonstrate that BFANet achieves consistent improvements over the DeepLabV3 + baseline with only a negligible increase in model complexity. Specifically, BFANet improves mIoU from 77.57% to 80.23%, mPA from 86.36% to 88.05%, and mDice from 87.52% to 88.78%. Moreover, the boundary-level evaluation further shows that BFANet improves Boundary F1 from 57.28% to 61.55% and Boundary IoU from 48.74% to 51.42%, indicating better preservation of fine object contours and more accurate boundary alignment. Meanwhile, the parameter count increases only from 5.82M to 5.84M, and GFLOPs increase from 53.03 to 53.75. These results show that BFANet provides a better balance among segmentation accuracy, boundary quality, and lightweight efficiency, especially for small objects, thin structures, and boundary-sensitive categories.
In the lightweight semantic segmentation network, a large receptive field and accurate feature fusion are crucial. However, the traditional dilated convolution is sparse for pixel category inference, and there is a feature mismatch when fusing different levels of features. To address these issues, we proposed a novel m...
Wen-Hao Wang, Lin-Hong Dai, Meng-Ke An et al.· IEEE Signal Processing Lette...· 0 citations
Few-shot segmentation (FSS) remains a significant challenge due to the scarcity of annotated data and the need for precise object localization across novel classes. Existing approaches often rely on single-backbone architectures and coarse priors, which struggle to capture detailed semantics and precise spatial alignme...
M. Ajmal, Guo-Hua Geng, Xiao-Feng Wang· International Conference on...· 0 citations
Experimental results demonstrate that the method significantly narrows the performance gap in data-constrained scenarios, achieving a mean Intersection over Union increase of 4.6% in one-shot and 5.5% in five-shot settings over the baseline.
Jin-Ming Guo, Li-Hsuan Chen, Yi-Chong Zeng et al.· APSIPA Transactions on Signa...· 0 citations
A Prior-Guided Part Aggregator (PGA) is designed, which leverages the foreground prior provided by foundation models to guide discriminative part discovery, enhancing target region responses while suppressing background interference, and a Topology-Informed Semantic Graph Convolutional Network (TIS-GCN) is designed to...
Xue-Rong Liu, Min Zhi, Yan-Jun Yin et al.· Journal of Imaging· 0 citations
A complementary prototype representation framework is proposed, employing three modules to collaboratively improve pseudo-label quality and improves the discriminative ability of confused categories by generating semantically similar sub-category negative samples.
RGB-D semantic segmentation has made notable progress by fusing RGB and Depth, yet mainstream models still learn features almost exclusively from pixel-level supervision, lacking direct high-level semantic constraints. This raises a central question-can external knowledge such as language priors inject stronger semanti...
Zi-Chuan Wei, Yin-Long Liu, Yan Xia et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.