Hongwei Zhang

In medical image segmentation, inherent boundary ambiguity, tissue overlap, and weak intensity gradients often produce blurred or discontinuous edges, posing persistent challenges to accurate anatomical delineation. Although deep learning approaches have advanced segmentation performance, their capacity to model ambiguous boundaries remains limited, especially under scarce annotation conditions. To address this, we propose SBEM-UNet, a novel semisupervised learning framework that explicitly targets boundary blurring and discontinuity through two complementary components: the semantic boundary enhancement module (SBEM) and the contour enhancement decoder (CED). SBEM leverages multiscale semantic aggregation with attention mechanisms to enhance structural discriminability and semantic coherence, mitigating boundary-induced uncertainty and preserving fine-grained details, while CED performs fine-grained contour modeling via dynamic boundary extraction and adaptive global feature modulation, enabling precise localization of blurred or discontinuous anatomical edges. To further improve robustness under limited annotation, the framework incorporates pseudolabel consistency regularization within the semisupervised learning paradigm, facilitating effective feature self-calibration and generalization across diverse imaging conditions. Experiments on public benchmarks demonstrate that SBEM-UNet consistently surpasses existing semisupervised methods in both region-based accuracy and boundary delineation quality, highlighting its effectiveness and practical value in low-annotation clinical scenarios.

Next Point-of-Interest (POI) recommendation is a crucial task in personalized location-based services, aiming to predict the next POI that a user might visit based on their historical trajectories. Although sequence models and Graph Neural Networks (GNNs) have achieved significant success, they often overlook the diversity and dynamics of user preferences. To address these issues, researchers have begun to employ Hypergraph Convolutional Networks (HGCNs) for disentangled representation learning. However, two critical problems have received less attention: (1) the limited expressive capacity of conventional hypergraph convolution layers, which restricts the modeling of complex nonlinear user–POI preference interactions and consequently weakens generalization performance, and (2) the inadequate utilization of contrastive learning mechanisms, which prevents fully capturing cross-view collaborative signals and limits the exploitation of complementary multi-view information. To tackle these challenges, we propose a Nonlinear Enhanced Disentangled Contrastive Hypergraph Learning (NE-DCHL) for next POI recommendation. The proposed model enhances nonlinear modeling capability and generalization by integrating ReLU activation, residual connections, and dropout regularization within the hypergraph convolution layer. A K-Nearest Neighbor (KNN)-based weighted adjacency matrix is employed to construct the geographical-view hypergraph, reducing computational complexity while maintaining essential spatial correlations. Moreover, a mini-batch InfoNCE loss and the GRACE (deep GRAph Contrastive rEpresentation learning) framework are utilized to improve efficiency and cross-view collaboration. Extensive experiments on two real-world datasets demonstrate that NE-DCHL consistently outperforms the original DCHL and other state-of-the-art approaches.

Inspired by the spores controlling during the life cycle of pyrosomes, this paper proposes a novel swarm intelligence-based meta-heuristic adaptive pyrosome optimization algorithm named APOA, which consists of four stages: Initialization, Information Interaction, Adaptive Decision-making, and Reaction. During the interaction, spores broadcast their information to each other. The decision-making stage employs a mechanism using random and adaptive strategies to enhance global exploration and local exploitation, respectively. Comprehensive tests on 23 test functions demonstrate the superior convergence and exceptional local optima avoidance of APOA. Extensive experimental results across CEC2014, CEC2017, CEC2020, and CEC2022 standard test suites reveal that APOA statistically outperforms 14 state-of-the-art metaheuristics in convergence speed, solution quality, and algorithmic stability. Furthermore, Empirical results on four types of engineering design problems further confirm that APOA is significantly superior to the leading competing algorithms in solving complex high-dimensional optimization challenges.

POI data play an important role in various location-based services, including navigation, positioning, and local search applications. However, as cities rapidly develop, a substantial amount of new POI data are generated daily, often accompanied by issues with the quality of their labels. Therefore, there is an urgent need to implement intelligent inference and enhancement processing for POI data labels. Conventional neural network models primarily target balanced data distribution, but they fail to address the issue of imbalanced distribution of POI data labels in terms of quantity. Furthermore, most neural network classification models implicitly learn the semantic knowledge of different categories from training datasets, neglecting the explicit semantic information offered by natural language labels. Considering the above problems, several negative samples are introduced for each input to a positive class, thereby transforming the multi-classification task into a binary classification problem. Simultaneously, POI data labels are introduced to provide explicit semantic information, and the semantic relationship between POI data labels and their names is determined using cross-coding. Experiments demonstrate that the macro − F1 score for the test dataset, which contains 75 different categories of POI data, reaches 0.84. This result surpasses the performance of traditional methods, highlighting the effectiveness of the proposed method.

The estimation of spatially resolved near-surface air temperature (NSAT) has been extensively performed in previous studies using satellite-derived land surface temperature (LST) from MODIS. However, there remains a need for estimating daily NSAT based on LST data from other satellites, which has important implications for integrating multi-source LST in estimating NSAT and ensuring the continuity of satellite-derived estimates of NSAT over long-term periods. In this study, we conducted a comprehensive comparison of LST derived from Metop with MODIS LST in the modeling and mapping of daily NSAT. The results show that Metop LST achieves consistent predictive performance with MODIS LST in estimating daily NSAT, and models based on Metop LST or MODIS LST have overall predictive performance of about 1.2–1.4 K, 1.5–2.0 K, and 1.8–1.9 K in RMSE for estimating Tavg, Tmax, and Tmin, respectively. Compared to models based on nighttime LST, daytime LST can improve the predictive performance of Tmax by about 0.26–0.28 K, while performance for estimating Tavg or Tmin using different schemes of LST is comparable. Models based on Metop LST also exhibit high consistency with models utilizing MODIS LST in terms of the variability in predictive performance across months, with RMSE of 1.03–1.82 K, 1.3–2.49 K, and 1.26–2.66 K for Tavg, Tmin, and Tmax, respectively. This temporal variability in performance is not due to sampling imbalance across months, which is confirmed by comparing models trained using bootstrapped samples in balance, and our results imply that sampling representativeness, complicated by retrieval gaps in LST, is an important issue when analyzing the variability in predictive performance for estimating NSAT. To fully assess the predictive capability of Metop LST in estimating daily NSAT, more studies need to be performed using different methods across areas with a range of scales and geographical environments.

Carbon compensation can guide human activities in reducing carbon emissions or increasing carbon sequestration and also represents an important approach for coordinating regional development. In this paper, Chongqing Municipality, whose internal development is varied, was selected as a case study. The annual carbon emissions, carbon sequestration, carbon deficits, and inter-regional carbon compensation costs from 2000 to 2021 were continuously estimated via local optimization methods, and a carbon compensation zoning scheme was proposed that integrates the present situation and trend analysis. The results show that (1) Chongqing’s total carbon emissions were greater than the total carbon sequestration, and the carbon deficit was approximately 556.24 × 10⁴ t~3621.58 × 10⁴ t. (2) County-level carbon budgets have large regional differences; the counties that should always receive carbon compensation are from the southeast and northeast regions, and the counties that should always pay carbon compensation are from central urban areas and the surrounding new urban areas. (3) All the counties were zoned into key payment areas, basic payment areas, key recipient areas, and basic recipient areas. The key payment areas, which account for 39.47%, maintain and grow payment status and are the main sources of carbon compensation costs, while the key recipient areas, which account for 44.74%, maintained a negative compensation status and a continuous downward trend, meaning that they may receive increasing carbon compensation costs. This paper revealed inequities in carbon compensation and proposed a novel zoning solution, which can provide scientific reference and data support for further establishing inter-regional carbon compensation mechanisms.