Selected work

Representative Papers

Enhancing generalization in Sickle Cell Disease diagnosis through ensemble methods and feature importance analysis

Feb 01, 2025Engineering applications of artificial intelligence

This study addresses the limited generalization and insufficient interpretability of existing diagnostic models for sickle cell disease by proposing an ensemble learning approach that integrates Random Forest and Extremely Randomized Trees. The method incorporates image preprocessing, red blood cell segmentation, and morphological feature extraction to construct a highly generalizable and interpretable diagnostic support system. A key feature identification mechanism is designed to reduce model complexity while significantly enhancing performance. Evaluated on a newly curated validation set, the proposed model achieves an F1-score of 90.71% and an SDS-score of 93.33%, outperforming current gradient boosting methods. The authors further promote reproducibility by open-sourcing the code, hyperparameters, and raw results.

2 citations1 influentialRead paper

Odometer-Agnostic Drift Correction Using OpenStreetMap Lane Geometry

Sep 09, 2026

本文提出了一种利用OpenStreetMap车道几何信息来校正长期漂移问题的轻量级方法，适用于各种里程计且无需密集地图或复杂预处理。

0 citationsRead paper

Hybrid multi-objective evolutionary algorithms for service placement in the computing continuum: a comparative study with genetic traceability

Jul 14, 2026

This study addresses the conflicting objectives of performance, resource utilization, and latency in multi-objective service placement across the edge–fog–cloud continuum by proposing a hybrid island-model-based multi-objective evolutionary algorithm. The approach co-evolves two heterogeneous subpopulations that periodically exchange solutions and introduces an innovative traceable analysis mechanism based on genetic load to quantify each sub-algorithm’s contribution to the final Pareto front. Integrated with NSGA-II, NSGA-III, U-NSGA-III, SMS-EMOA, MOEA/TS, and MOCPO, the proposed method demonstrates statistically significant superiority over individual algorithms across 30 independent runs. Comprehensive evaluation using metrics including Generational Distance (GD), Inverted Generational Distance (IGD), Hypervolume (HV), Spacing (S), and Set Coverage with Traceable Explanation (STE) confirms its advantages in scalability, optimization efficacy, and interpretability.

0 citationsRead paper

Attention-Based Segmentation of WMHs and Differentiation of Vascular vs. Demyelinating Lesions

Jul 09, 2026

This study addresses the diagnostic challenge posed by the similar imaging appearances of white matter hyperintensities (WMHs) in vascular and demyelinating diseases. To improve differential diagnosis, the authors propose an integrated segmentation-and-classification framework. First, they achieve precise WMH segmentation using an Attention U-Net enhanced with BAM/CBAM attention mechanisms, 2.5D input representation, and a patch-based training strategy. Subsequently, morphological features derived from the segmentation masks are employed to drive lesion-type classification. This work is the first to synergistically combine these attention modules with a 2.5D patch approach for WMH segmentation. The method demonstrates significantly improved generalization and etiological discrimination performance across five multi-protocol public MRI datasets.

0 citationsRead paper

Variational Deep Unfolding with Mamba-Based Nonlocal Modeling for Underwater Image Enhancement

Jun 10, 2026

This work addresses the challenges of low visibility and color distortion in underwater images caused by light scattering and absorption by proposing a variational model-based deep unfolding network. For the first time, the Mamba architecture is introduced into underwater image enhancement, leveraging a dehazing decomposition, multiplicative residual terms, and non-local gradient constraints to model scene self-similarity. A proximal trajectory loss is specifically designed to ensure consistency between the unfolding process and ideal regularized iterations. The proposed framework offers both theoretical guarantees on solution existence and strong capability in fine detail recovery, significantly outperforming state-of-the-art methods in both visual quality and quantitative metrics, thereby effectively enhancing image clarity and color fidelity.

0 citationsRead paper

Recent publications

Latest Papers

Odometer-Agnostic Drift Correction Using OpenStreetMap Lane Geometry

Sep 09, 2026

本文提出了一种利用OpenStreetMap车道几何信息来校正长期漂移问题的轻量级方法，适用于各种里程计且无需密集地图或复杂预处理。

0 citationsRead paper

Hybrid multi-objective evolutionary algorithms for service placement in the computing continuum: a comparative study with genetic traceability

Jul 14, 2026

This study addresses the conflicting objectives of performance, resource utilization, and latency in multi-objective service placement across the edge–fog–cloud continuum by proposing a hybrid island-model-based multi-objective evolutionary algorithm. The approach co-evolves two heterogeneous subpopulations that periodically exchange solutions and introduces an innovative traceable analysis mechanism based on genetic load to quantify each sub-algorithm’s contribution to the final Pareto front. Integrated with NSGA-II, NSGA-III, U-NSGA-III, SMS-EMOA, MOEA/TS, and MOCPO, the proposed method demonstrates statistically significant superiority over individual algorithms across 30 independent runs. Comprehensive evaluation using metrics including Generational Distance (GD), Inverted Generational Distance (IGD), Hypervolume (HV), Spacing (S), and Set Coverage with Traceable Explanation (STE) confirms its advantages in scalability, optimization efficacy, and interpretability.

0 citationsRead paper

Attention-Based Segmentation of WMHs and Differentiation of Vascular vs. Demyelinating Lesions

Jul 09, 2026

This study addresses the diagnostic challenge posed by the similar imaging appearances of white matter hyperintensities (WMHs) in vascular and demyelinating diseases. To improve differential diagnosis, the authors propose an integrated segmentation-and-classification framework. First, they achieve precise WMH segmentation using an Attention U-Net enhanced with BAM/CBAM attention mechanisms, 2.5D input representation, and a patch-based training strategy. Subsequently, morphological features derived from the segmentation masks are employed to drive lesion-type classification. This work is the first to synergistically combine these attention modules with a 2.5D patch approach for WMH segmentation. The method demonstrates significantly improved generalization and etiological discrimination performance across five multi-protocol public MRI datasets.

0 citationsRead paper

Variational Deep Unfolding with Mamba-Based Nonlocal Modeling for Underwater Image Enhancement

Jun 10, 2026

This work addresses the challenges of low visibility and color distortion in underwater images caused by light scattering and absorption by proposing a variational model-based deep unfolding network. For the first time, the Mamba architecture is introduced into underwater image enhancement, leveraging a dehazing decomposition, multiplicative residual terms, and non-local gradient constraints to model scene self-similarity. A proximal trajectory loss is specifically designed to ensure consistency between the unfolding process and ideal regularized iterations. The proposed framework offers both theoretical guarantees on solution existence and strong capability in fine detail recovery, significantly outperforming state-of-the-art methods in both visual quality and quantitative metrics, thereby effectively enhancing image clarity and color fidelity.

0 citationsRead paper

Analysis of wireless network access logs for a hierarchical characterization of user mobility

May 14, 2026

This study addresses the challenges of modeling user mobility in large-scale Wi-Fi logs, particularly the trade-off between model complexity and geographic adaptability. The authors propose a hierarchical modeling approach that constructs user trajectories from sequences of Wi-Fi access points, recursively clusters geographic features to derive multi-granular spatial regions, and integrates user profiles to learn state transition matrices and dwell-time vectors. Evaluated on a campus dataset from the University of the Balearic Islands, the method significantly reduces model complexity while preserving accuracy and enhancing generalization across diverse geographic contexts. Experimental results demonstrate that the hierarchical approach substantially outperforms non-hierarchical baselines in modeling transition dynamics, although dwell-time prediction remains an area requiring further refinement.

0 citationsRead paper