Hybrid multi-backbone architectures and the utilization of edge cues for auxiliary training have become two major research trends in salient object detection (SOD). It is widely acknowledged that CNNs can effectively model local spatial structures, while Transformers can capture long-range global dependencies. However, the representation discrepancy between CNN and Transformer features, together with boundary-detail degradation during multi-scale fusion, remains a major challenge. In addition, how to effectively leverage edge cues as reliable structural guidance without introducing texture-induced false boundaries or boundary leakages remains an open issue. In this paper, we present SECA-Net, a unified framework that establishes a profound synergy between CNN and Transformer representations. It explicitly bridges their inherent discrepancies through level-dependent interaction strategies, while resolving structural degradation via a sequential "purify-and-guide" mechanism. This approach enables the network to extract and utilize edge cues effectively, thereby alleviating boundary degradation and texture-induced false contours. Specifically, we design a dual-encoder structure to extract features. A level-wise feature interaction (LFI) module is introduced to perform discrepancy-aware fusion across feature levels, stabilizing CNN-Transformer aggregation. Meanwhile, the features extracted from the CNN branch are projected into a semantic-aware edge refinement (SAER) module to produce clean multi-scale edge priors under high-level semantic guidance, suppressing texture-induced spurious edges. Finally, we design an edge-guided cross-attention feature aggregation (ECFA) module, which progressively injects refined edge priors as structural constraints into multi-scale saliency decoding via cascaded cross-attention, enabling effective structural refinement. Overall, LFI reduces cross-branch discrepancy, SAER purifies boundary priors, and ECFA integrates semantics and structure in a progressive decoding manner, forming a unified SECA-Net framework. Extensive experimental results on five benchmark SOD datasets show that SECA-Net outperforms 19 state-of-the-art methods, demonstrating its effectiveness. Specifically, our proposed method ranks first in Fβ and BDE across all datasets, notably improving Fβ by 1.54% on the challenging DUTS-TE dataset.
Intraspecific variation in phenotypic plasticity can affect the ability of populations, and thus species, to respond to environmental changes. However, the prevalence and drivers of such variation are not well known. Most proposed explanations for intraspecific variation in phenotypic plasticity involve mechanisms associated with a population's position within the species' geographic range or the environmental heterogeneity experienced by the population. To assess the effect of these two drivers, and their potential interaction, we use a combination of germination and greenhouse experiments to measure thermal phenotypic plasticity in traits ranging from germination probability to flower abundance in populations of three Hypericum species sampled across their European ranges. We then relate thermal plasticity to each population's position within the species' range and to the environmental heterogeneity of the sampling site. Our results revealed that, while average thermal plasticity in several traits was similar among the three tested Hypericum species, it varied among the conspecific populations. Specifically, populations closer to the range edge tended to be more plastic in germination probability and plant height, while populations from more heterogeneous environments tended to be more plastic in flowering phenology, plant height, and flower abundance. Interestingly, for plasticity in germination phenology, plant height, and flower abundance, we found a substantial interactive effect with accentuated plasticity in heterogeneous sites near the range edge. This suggests that populations in heterogeneous environments at range edges may adjust to environmental change via phenotypic plasticity more effectively than do other conspecific populations. These results support both tested drivers and reveal important interactive patterns for some of the tested traits. Furthermore, they encourage further research on plasticity that considers both range position and environmental heterogeneity.
Transcatheter edge-to-edge repair (TEER) has become an effective alternative for treating degenerative mitral regurgitation (DMR) in patients at high surgical risk. The SQ-Kyrin-M TEER system (SQ-Kyrin-M system) is a novel TEER device developed in China. This study aimed to evaluate the feasibility, safety, and 12-month clinical efficacy of the SQ-Kyrin-M system in patients with high-risk degenerative mitral regurgitation. In this prospective, multicenter, single-arm study (ClinicalTrials.gov: NCT06467110), 120 patients with symptomatic DMR (grade ≥3+) had the device implanted. The primary endpoint was the clinical success rate at 12 months. Secondary endpoints included technical, device, and procedural success rate; New York Heart Association (NYHA) class improvement; Kansas City Cardiomyopathy Questionnaire (KCCQ) score change; and mitral regurgitation (MR) reduction. Safety endpoints encompassed all-cause mortality, cardiovascular mortality, and major adverse event rate. A total of 120 patients received the TEER procedure across 25 participating sites in China; the mean age was 71.9 years, and the mean Society of Thoracic Surgeons (STS) risk score was 9.3. At 12 months, the Kaplan-Meier estimates were 82.5% for clinical success, 7.6% for all-cause mortality, and 10.8% for major adverse events; MR ≤2+ and MR ≤1+ were achieved in 91.7 and 70.4% of the patients, respectively; 88.9% of patients were in NYHA class I or II; and KCCQ score had improved by 18.9 points. Favorable left ventricular remodeling was observed with sustained reductions in left ventricular end-diastolic and end-systolic volumes. This study demonstrates that the SQ-Kyrin-M system is a safe and effective therapeutic option for DMR patients.
The widespread adoption of large-scale sensor networks in privacy-sensitive and safety-critical applications has intensified the demand for secure, trustworthy, and energy-efficient learning mechanisms at the network edge. Federated learning has emerged as a promising paradigm for privacy preservation by enabling collaborative model training without sharing raw sensor data. However, most existing federated approaches inadequately address trust management, communication efficiency, and energy constraints, which are critical in real-world sensor-based systems. This paper proposes a trust-aware and energy-efficient federated learning framework specifically designed for secure sensor networks operating in resource-constrained edge environments. The proposed approach integrates lightweight trust metrics, trust-driven model aggregation, and adaptive communication scheduling to mitigate the impact of unreliable or malicious nodes while reducing unnecessary energy expenditure. By dynamically weighting client contributions based on trust and participation efficiency, the framework enhances robustness and learning stability under heterogeneous sensing conditions. Experimental results show that the proposed method maintains significantly higher accuracy under adversarial participation while reducing communication overhead and cumulative energy consumption. In particular, the framework improves model accuracy by up to 3.2% under heterogeneous conditions, reduces communication overhead by 28%, and decreases cumulative energy consumption by 31% compared with conventional federated learning approaches.
Tricuspid valve surgery for tricuspid regurgitation (TR) following prior valve surgery carries increased risk. This study evaluated 2-year outcomes of tricuspid transcatheter edge-to-edge repair (T-TEER) in patients with severe TR and prior valve surgery. TRILUMINATE Pivotal is an international randomized trial comparing T-TEER with the TriClip (device group) to medical therapy (control group) in patients with symptomatic severe TR, with a concurrent single-arm cohort (device) for anatomically complex patients. Echocardiograms were assessed in a core laboratory, and outcomes were adjudicated by an independent clinical events committee. Among the 469 patients in the device group, 113 had prior valve surgery and 356 did not. Baseline characteristics were comparable in the 2 groups: mean age, 77 years versus 79 years; female sex, 64% versus 59%; atrial fibrillation, 87% versus 86%; and heart failure hospitalization (HFH) within 1 year prior to T-TEER, 26% versus 25%. The T-TEER success rate was 99% in both groups, with no in-hospital deaths, a median hospital stay of 1 day, and 97% discharged home. Thirty-day adverse event rates were low and similar: rates of all-cause mortality, stroke, and new pacemaker implantation were all <2%, and the rate of major bleeding was <4%. At 2 years, outcomes remained favorable and comparable: ≤ moderate TR was achieved in 81% versus 80% (P = .93), New York Heart Association class I/II was observed in 74% versus 84% (P = .11), and KCCQ scores improved by a mean of 15 ± 21 points versus 16 ± 23 points (P = .66). Both groups experienced a significant reduction in HFH (79% vs 31%; P = .01). In patients with prior valve surgery, T-TEER was safe and resulted in significant TR reduction and symptom improvement.
The Roman province of Dacia, located north of the Danube frontier, represented a key zone of cultural and demographic interaction during the Imperial period. However, the biological impact of Roman colonization in this region has not been characterized using genomic data. Here, we analyze genome-wide data from 34 individuals recovered from the Apulum- Dealul Furcilor necropolis, one of the largest funerary complexes in Roman Dacia. The genome-wide data reveal pronounced genetic heterogeneity within this population, reflecting its position at the intersection of Eastern Europe, the Mediterranean, and West Asia. Notably, we observe a sex-biased pattern of ancestry. Female individuals show stronger affinities to Eastern European, Steppe, and Caucasus-associated populations, suggesting the persistence of local or regionally connected genetic lineages. In contrast, male individuals display closer genetic relationships with Mediterranean and North African groups, including populations associated with Roman and Punic contexts, indicating male-mediated gene flow linked to long-distance mobility. These findings highlight the complex demographic processes shaping Roman frontier communities, where local and incoming populations were integrated through asymmetric social dynamics. Our results provide genomic evidence consistent with sex-biased admixture in Roman Dacia and underscore the role of frontier regions as hubs of genetic and cultural interaction within the Roman Empire.
Current research in gas sensor technology emphasizes developing high-performance, miniaturized devices that operate at room temperature. Among emerging materials, molybdenum disulfide (MoS2), a layered semiconductor, has garnered significant attention for its ability to detect diverse analytes with a high surface to volume ratio. In this study, composites of 1T and 2H MoS2 with reduced graphene oxide (rGO) were synthesized, combining distinct physical and chemical properties that enable unique interactions with gas molecules. The applied synthesis routes are cost effective, reproducible, and readily compatible with field effect transistor printed devices. Sensor performance for nitric oxide (NO), nitrogen dioxide (NO2), and ammonia (NH3) gases in 1T-based hybrids was superior to that in 2H-based hybrids, which was attributed to the metallic and hydrophilic nature of the 1T phase. The hybrids displayed excellent performance across a wide concentration ranging from 500 ppb to 2 ppm. Notably, for NO detection, the 1T MoS2@rGO sensor achieved responses of 5.4% at 500 ppb and 38.1% at 2 ppm. Density functional theory further confirmed the metallic character of the 1T phase. These results underscore the promise of phase-engineered MoS2@rGO hybrids as next-generation materials for reliable room temperature gas sensing technologies.
Background/Objectives: Wound management presents a substantial clinical challenge due to the rising incidence of chronic wounds, infections, and the limitations of conventional dressings in creating an ideal healing microenvironment. This review aims to provide a comprehensive overview of advanced smart hydrogel platforms designed to improve wound healing outcomes, focusing on their capacity to respond adaptively to physiological and external stimuli. Methods: This article analyzes the core characteristics of smart hydrogels, specifically examining stimuli-responsive systems (pH, temperature, enzyme, light, and electricity). The review evaluates advanced configurations-including injectable, self-healing, and 3D-printable systems-and functionalized hydrogels integrated with antimicrobials, drugs, and nanocomposites. Additionally, essential characterization methodologies, biological assessments, and regulatory considerations for clinical translation are synthesized. Results: The literature, which is predominantly preclinical in nature, indicates that functionalized hydrogels significantly enhance tissue regeneration, angiogenesis, and infection control compared to traditional methods. Conductive hydrogels utilizing bioelectrical signals show particular promise in accelerating the healing process. While current clinical applications and commercial products demonstrate efficacy, significant barriers remain regarding large-scale manufacturing and regulatory approval. Conclusions: Smart hydrogels represent a transformative approach to precision wound management, offering superior adaptability and therapeutic delivery. To achieve widespread clinical adoption, future research must address manufacturing scalability and focus on emerging trends, such as the integration of biosensors and AI-powered monitoring systems, to create fully intelligent wound care solutions.
Maternal anger, an intense and often stigmatized emotional experience, has been associated with postpartum mood and anxiety disorders. This biomedical lens obscures its relational and sociocultural roots and remains understudied in psychological and sociological literature, especially from a qualitative stance contextualised in an LMIC. Drawing on Benjamin's feminist psychoanalytic theory and Bronfenbrenner's social-ecological model, this study explores maternal anger as a meaningful, biopsychosocial response to relational ruptures and systemic inequities in an Indian context. Vignettes of two urban Indian mothers from diverse backgrounds were chosen to present their lived experiences with maternal anger. Their in-depth, semi-structured interviews were analyzed using the Interpretive Phenomenological Approach. A relational-intersubjective lens guided attention to mutual recognition/ruptures and affective attunement/misattunement, while the social-ecological model mapped influences across the nested systems. The first theme that emerged was Experience of Anger in Relational Qualms, which included the sub-themes of Betrayal and Abandonment in Intimate Relationships and Erasure of their Personhood. The second theme, Socio-cultural Experience of Anger, consisted of Emotional Burden of the Care and Lack of Support, Suppression and Silence, and Intergenerational Scripts. Reconceptualising maternal anger at the intersection of interpersonal and systemic pressures reframes it from pathology to protest. It becomes an embodied critique of relational, gendered and systemic inequities and calls for perinatal mental health interventions that honor mothers' emotional realities. By depathologizing maternal anger, the lived experiences of the two urban Indian mothers contribute novel insights to maternal mental health and underscore the necessity of feminist, context-sensitive approaches in research and care.
Expert-guided Causal Structure Learning (CSL) incorporates prior knowledge to improve the accuracy of causal discovery, yet the acquisition of such knowledge is often restricted by the availability of human experts. While Large Language Models (LLMs) provide an alternative source of causal priors, LLM-derived knowledge can be inconsistent with the true causal structure due to hallucinations or contextual misinterpretations. This paper introduces a structural constraint measurement framework, which defines constraint strength and constraint quality to describe reliability and effectiveness, enabling a systematic evaluation of LLM-derived constraints. Using this framework, we evaluate five categories of structural constraints: Edge Existence (EEC), Edge Forbidden (EFC), Path Existence (PEC), Path Forbidden (PFC), and Order Constraints (OC). Our theoretical and empirical analyses demonstrate that while EEC offers high constraint strength, it exhibits low quality when derived from LLMs; conversely, PFC and OC provide a balanced trade-off between search-space pruning and reliability. Building on these insights, we propose a two-level CSL optimization framework that partitions the search space by node order and refines the structure using global path constraints. The results show that this framework provides an effective way to incorporate noisy LLM-derived priors into CSL, particularly in settings where expert knowledge is limited.
Elevator fault diagnosis heavily relies on high-precision sensing of microscopic physical states. Although Micro-Electro-Mechanical System (MEMS) sensors can capture such subtle features, they are constrained by high-frequency data streams, environmental noise, and the semantic gap between raw sensor data and actionable maintenance decisions. This study proposes a collaborative edge-cloud intelligent diagnosis framework specifically designed for elevator systems. On the edge side, a lightweight temporal Transformer model, ELiTe-Transformer, was designed and deployed on the Jetson platform. This model enhances sensitivity to event-driven MEMS signals through an industrial positional encoding mechanism and by integrating linear attention and INT8 quantization techniques, achieving a real-time inference latency of 21.4 ms. On the cloud side, retrieval-augmented generation (RAG) technology was adopted to integrate physical features extracted at the edge with domain knowledge, generating interpretable diagnostic reports. The experimental results show that the overall accuracy of the system reaches 96.0%. The edge-cloud collaborative framework improves the accuracy of complex fault diagnosis to 92.5%, and the adoption of RAG reduces the report hallucination rate by 71.4%. This work effectively addresses the bottlenecks of MEMS perception in elevator fault diagnosis, forming a closed loop from micro-signal acquisition to high-level decision support.
BackgroundDetecting breast cancer, especially identifying microcalcifications in mammograms, is challenging due to the need for high sensitivity and efficient processing. This study presents a novel algorithm, Sigmoidal Slope Analysis and Aspect Ratio Evaluation (SAAR), designed for real-time application on edge devices. By employing a multi-step adaptive process with sigmoidal functions, SAAR enhances intensity contrast and prioritizes regions of interest, enabling fast, accurate detection of microcalcifications.ObjectiveThis study aims to develop and validate an efficient, edge-device-compatible method for detecting microcalcifications in mammographic images. The goal is to provide a tool that enhances diagnostic efficiency through real-time processing, thereby supporting early breast cancer detection in both clinical and remote settings.MethodsThe SAAR algorithm utilizes an adaptive slope detection technique based on the sigmoid function, dynamically adjusting to local intensity features. This approach allows for greater adaptability to image variations. The algorithm prioritizes regions of interest through a multi-step adaptive process, enhancing intensity differences to focus on potential microcalcifications.ResultsTesting on established mammography databases, such as MIAS, demonstrates the algorithm's effectiveness, with improved sensitivity compared to conventional methods. Designed for edge devices, the algorithm leverages their real-time processing capabilities, offering lower latency and enhanced privacy.ConclusionsThe integration of SAAR with edge devices represents a promising advancement in breast cancer detection. The adaptive nature of SAAR, coupled with the real-time processing capabilities of edge devices, provides a robust solution for enhancing microcalcification detection efficiency and sensitivity in mammography.
Detecting surface defects on steel products is crucial for maintaining quality standards in industrial manufacturing. However, existing detection algorithms face several challenges, including the difficulty of capturing multi-scale defect characteristics with fixed receptive fields, insufficient utilization of defect edge and frequency domain features, and simplistic feature fusion strategies. In response to the above challenges, this paper proposed the Multi-Scale Frequency-Enhanced YOLO (MSFE-YOLO) algorithm that integrates multi-scale frequency domain enhancement with defect-aware attention mechanisms. First, a Multi-Scale Frequency-Enhanced Convolution (MSFC) module was constructed, which extracted multi-scale spatial features in parallel through depth-adaptive dilated convolutions, explicitly modeled high-frequency edge information using the Laplacian operator, and achieved adaptive fusion of multi-branch features via learnable weights. Second, a Cross-Stage Partial with Multi-Scale Defect-Aware Attention (C2MSDA) module was designed, integrating Sobel operator-based edge perception, multi-scale spatial attention, and adaptive channel attention to collaboratively enhance features across spatial, channel, and edge domains through a gated fusion strategy. Finally, an Adaptive Feature Fusion Enhancement (AFFE) module was proposed to achieve adaptive aggregation of multi-level features through a data-driven weight generation network and cross-scale feature interaction mechanism. Experimental results on the NEU-DET and GC10-DET datasets demonstrated that MSFE-YOLO achieved the mAP@0.5 of 79.8% and 66.7%, respectively, which were 1.7% and 2.1% higher than the benchmark model YOLOv11s respectively, while maintaining an inference speed of 89.3 FPS, which satisfied the real-time detection requirements in industrial scenarios.
In IIoT edge computing, multi-edge server collaborative scheduling faces two core issues due to random task arrivals, heterogeneous resources, and complex topology: traditional model-driven methods cannot make dynamic decisions in dynamic environments, and conventional MARL fails to characterize inter-node topological dependencies and load correlations. To address this, this paper investigates the joint optimization of task offloading, computing resource allocation, and SFC orchestration in IIoT, constructs a cloud-edge-end collaborative architecture, and models the problem as a POMDP to minimize the overall system cost under multiple constraints. A graph-guided value-decomposition MARL method is proposed, which extracts spatial topology and neighborhood-load features of edge nodes via a GNN and combines them with the QMIX framework to realize multi-agent centralized training and distributed execution. Simulations show that the algorithm converges stably under different server scales and task loads, significantly outperforms benchmark algorithms, and can suppress performance degradation in high-load scenarios, demonstrating its robustness and scalability in complex industrial environments.
Under operando conditions, how local structural distortion influences electrocatalysis in non-metallic materials remains poorly understood, largely because charge-transfer processes and subsequent chemical steps are strongly coupled and difficult to disentangle experimentally. Here, by using operando atomic force microscopy-scanning electrochemical microscopy (AFM-SECM), we directly identify wrinkled regions in monolayer molybdenum disulfide (MoS2) as highly active domains for the hydrogen evolution reaction (HER). Combined local AFM-SECM and scanning transmission electron microscopy (STEM) imaging further reveal that the enhanced activity is primarily localized at wrinkle edges, where folded edge structures are formed, providing spatially resolved evidence of a local structure-activity relationship. Interestingly, operando electron-transfer (ET) imaging reveals only limited enhancement of charge-transfer kinetics in these regions, indicating that the increased HER activity more likely arises from the promotion of subsequent chemical steps rather than from improved electron transfer. These findings provide mechanistic insight into the role of folded edge structures within wrinkled regions in non-metallic electrocatalysis and offer guidance for the rational design of high-performance electrocatalytic materials.
Mangrove ecosystems are vulnerable to extreme events and sea-level rise. The present study examined how biological and geomorphological processes interact at mangrove seaward margins (MSM) using remote sensing, field surveys, and elevation monitoring from 2009 to 2021 in northeastern Hainan, China. The results showed that the mangrove edge had retreated by 11.04 ± 0.36 m. Retreat was more rapid in heavily disturbed areas, especially during 2013-2017 when intense tropical cyclones occurred. In these areas, edge seedling recruitment of Rhizophora stylosa declined, with establishment probability being strongly influenced by surface elevation. Mature trees exhibited increased root damage and reduced leaf chlorophyll content, with trait variation primarily driven by surface elevation and sediment physical properties. Following vegetation dieback, surface elevation declined rapidly, forming a positive biogeomorphic feedback that further inhibited regeneration and accelerated margin degradation. We identified a feedback loop between vegetation loss and geomorphic change triggered by extreme disturbances at the MSM that limits natural recovery and threatens ecosystem stability. Our findings underscore the need to prioritize monitoring edge zones and suggest a management framework that integrates remote sensing with in situ monitoring to identify vulnerable zones and guide post-disturbance conservation and restoration under increasing environmental stress.
In low-latency edge-intelligence scenarios such as autonomous driving and industrial edge analytics, the processing of large-scale sensor data imposes extremely stringent requirements on communication latency. However, the high overhead of the traditional TCP protocol makes it difficult to satisfy such demands, while the semantic gap between the high-performance RoCE protocol and the standard Socket API prevents existing applications from directly exploiting its advantages. To address this problem, this paper proposes TransBridge, a lightweight user-space communication middleware that transparently bridges TCP and RoCE. Its design is realized through three key innovations: a transparent user-space compatibility architecture that enables unmodified Socket-based applications to benefit from RoCE performance; a microsecond-level low-latency transmission engine that bypasses kernel and protocol stack overhead; and a lightweight lock-free resource management mechanism based on a decentralized peer-to-peer architecture and deferred buffer updates. Experiments on a real RoCE network show that TransBridge significantly outperforms mainstream schemes: it achieves an average round-trip latency of 5.926 μs for 16 B messages and a throughput of 20.254 Gbps for 16 KB messages; in the Fast DDS application-level evaluation, it achieves a throughput of 188 Mbps and an average round-trip latency of about 150 μs. The results indicate that TransBridge can provide transparent and effective RoCE acceleration for existing Socket-based applications in resource-constrained edge environments.
Heteroatom substitution is a promising strategy to enhance hydrogen evolution reaction (HER) activity of MoS2, yet synergistically activating both its basal plane and edge sites remains challenging. Herein, we report a dual-site substitution of both Mo and S with tellurium in the MoS2 lattice (Te-MoS2), which achieves a superior large-current-density HER performance in acidic electrolyte, surpassing all previously reported single-element-doped MoS2 with nonmetal or non-precious metal. The Te-MoS2 catalyst requires an overpotential of only 364 mV to achieve an industrial-level current density of 1000 mA·cm-2, significantly lower than 506 mV required by commercial 20 wt% Pt/C, and maintains this performance stably for 200 h without decay. Comprehensive analyses reveal that the simultaneous substitution of Mo and S with Te atoms activates neighboring S atoms and also promotes the formation of smaller, edge-rich MoS2 nanosheets, thereby generating abundant basal plane and edge S active sites with optimized hydrogen adsorption energy.
Forgetting is a fundamental component of adaptive memory and essential for cognitive flexibility, yet its cellular basis remains unclear. Here we establish a mouse model of retroactive interference (RI) and show that post-learning novelty exploration induces active forgetting of hippocampus-dependent object location memory only within a discrete consolidation window defined by protein synthesis sensitivity. RI imposed within this window reduces engram reactivation and destabilizes a structured coactivity network formed during learning. Mice that forget retrieve reorganized engram with increased edge turnover and reduced training edge survival during recall. During this vulnerable period, RI infiltrates the engram core, whereas after consolidation it remains confined to the network periphery. Over the consolidation window, the engram network progressively matures, acquiring greater density, similarity, and k-core robustness, features that confer resistance to interference. Importantly, blocking RI infiltration rescues memory formation. Together, these findings show that forgetting arises from reorganization of engram topology during consolidation and identify engram core contamination as a network-level substrate for forgetting.
Cryo-electron tomography (cryo-ET) enables in situ three-dimensional visualization of many protein complexes and other macromolecular assemblies such as ribosomes in cells, yet automated macromolecule particle identification in 3D cryo-ET tomograms remains a major bottleneck due to dose-limited low signal-to-noise ratios, missing-wedge artifacts, and densely crowded cellular backgrounds. We present TomoSwin3D, an end-to-end three-dimensional (3D) macromolecule particle identification and classification pipeline centered on a Swin Transformer-based U-Net that performs particle identification and classification and outputs particle centroid coordinates. TomoSwin3D leverages a multi-channel input representation that augments raw tomogram densities with complementary 3D feature maps capturing edge strength (Sobel gradients), local contrast enhancement (morphological top-hat), and multiscale blob responses (Difference-of-Gaussians), improving detectability of small and low-contrast targets. To better preserve particle geometry and avoid hand-crafted shape assumptions, it adopts occupancy-preserving supervision that directly uses available 3D instance masks rather than heuristic Gaussian/spherical labels and applies scalable patch-wise inference followed by lightweight post-processing (connected-component analysis, size filtering, centroid extraction) for robust centroid coordinate extraction. Across diverse simulated and experimental cryo-ET tomogram benchmarks including SHREC 2021 and 2020 test datasets, EMPIAR dataset, and Cryo-ET data portal dataset, TomoSwin3D achieves strong and consistent performance in detecting proteins and other particles, outperforming existing methods, with a pronounced advantage in picking hard, small protein particles. These results establish TomoSwin3D as a scalable and accurate solution for high-throughput cryo-ET macromolecule particle picking and downstream subtomogram averaging.