Hailiang Ye

Deep learning methods are becoming increasingly important for medical image segmentation. However, previous approaches often performed poorly in spatial-frequency-domain fusion and feature fusion, leading to inaccurate lesion localization, hazy boundaries, and low accuracy. This paper proposes a reliability-guided fusion and dynamic dual-domain mask-modulated Shearlet network for medical image segmentation (RF-DMSNet for short). Its two primary components are the reliability-guided multi-scale feature fusion module (RGMF) and the dynamic dual-domain mask with a coordinate attention-modulated Shearlet operator (DAMS). The former module exhibits reliability in both the channel and spatial dimensions, allowing adaptive weighted fusion to increase feature quality. Meanwhile, the latter uses a dynamic dual-domain mask and coordinate attention to improve lesion boundary and location segmentation performance. The frequency-domain re-enhancement module (FREM) and critical feature guided module (CFGM) are designed using DAMS. These two modules work together to optimize a unified, detailed enhancement module (DEM), enabling progressive refinement of segmentation predictions from coarse to fine. Extensive experimental results demonstrate that RF-DMSNet achieves superior segmentation performance compared to state-of-the-art approaches for medical image segmentation.

Reliable short-term freeway traffic forecasting is essential for proactive traffic management, including congestion warning, ramp metering support, and traveler information services. However, many existing forecasting models treat spatial dependencies as synchronous or weakly directional, limiting their ability to represent delayed upstream–downstream traffic propagation. This study proposes Traffic-DiMAGNet, a lightweight and interpretable lag-aware directed spatio-temporal graph neural network for real-time freeway flow forecasting. The model constructs a sparse directed sensor dependency graph by integrating physical road adjacency, training-set lead–lag traffic priors, and learnable source–target node embeddings. A lag-aware bidirectional propagation module then shifts inter-sensor messages according to estimated propagation delays, while directed random-walk normalization, directional gating, and multi-scale causal convolutions preserve asymmetric traffic semantics with low computational cost. Experiments on four Caltrans PeMS datasets show that Traffic-DiMAGNet consistently outperforms recurrent, diffusion-based, attention-based, and adaptive-graph baselines regarding MAE, RMSE, and MAPE values. The learned directed lag structures provide interpretable propagation information, and the lightweight architecture supports rolling 5 min forecasting, indicating practical potential for real-time freeway monitoring and proactive traffic management.

Deep learning has achieved remarkable success in point cloud completion. However, existing methods easily struggle with geometric variations, such as rotations, translations, and structural differences among points. Accordingly, this paper develops a geometric-aware framework integrated with novel sharpen-hypergraph convolution and equivariance operators (GA-SHE) for rigid point cloud completion, achieving more uniform and reliable results. This framework is built upon two key components: an equivariant feature extractor (EFE) in the encoder and a sharpen-hypergraph convolution-based coarse generation (SHCG) module in the decoder. The former constructs an equivariance convolution to learn equivariance feature representations to rotations and translations, effectively leveraging the intrinsic geometric structures of point clouds and contributing to point cloud completion. A theoretical analysis of the equivariance property of EFE is also provided. Moreover, in the SHCG module, a novel sharpen-hypergraph convolution operator is proposed, modeling the high-order structural difference relationships by encouraging the central node to diverge from its neighbors. Based on this operator, SHCG enables the network to effectively capture high-order geometric difference-aware features among points and generate more uniform point clouds. Experiments on various datasets demonstrate the effectiveness and competitive performance of GA-SHE. Especially, the results highlight the effectiveness of the sharpen-hypergraph convolution operator and confirm that the equivariance representations learned by EFE align with the theoretical analysis.