Search Results (93)

3D bird pose estimation plays a pivotal role in ecological conservation research. However, it remains a formidable challenge due to extensive joint deformation, severe self-occlusion, and the scarcity of 3D ground truth data. Therefore, practical solutions typically rely on accurate 2D keypoint detection from monocular images and subsequent 3D lifting. Although the High-Resolution Network (HRNet) has established a benchmark in 2D pose estimation by preserving high-resolution feature representations, its architecture, which relies on small convolution kernels, faces difficulties in capturing the global long-range dependencies necessary to resolve severe occlusions. To address these deficiencies, the core contributions of this work are summarized as follows: (1) We design a Gated LS-Block with a partial channel gating strategy to decouple channel mixing from spatial mixing, and extract global long-range dependencies via the proposed Large–Small Convolution (LSConv) to minimize feature redundancy. (2) We embed this block into Stage 2 of HRNet, enhancing multi-scale feature learning while slightly reducing model parameters and computational overhead; (3) To alleviate the ill-posed nature of monocular 3D lifting without paired supervision, we develop an unsupervised 3D reconstruction algorithm. Experimental results on the Animal Kingdom dataset demonstrate that our method achieves a 0.9% improvement in PCK@0.05 while reducing GFLOPs by 3.3%. These results verify that the proposed architecture enhances the model’s representation capability for bird poses while ensuring efficient inference. Meanwhile, we validate the applicability of the proposed 3D reconstruction algorithm via qualitative experiments, and further demonstrate that our unsupervised 3D lifting algorithm successfully preserves low symmetry error and robust bone length consistency with proxy metrics. Full article

China’s vast rural landscape exhibits pronounced regional disparities in both foundational resources and development potential. In the context of nationwide rural revitalization efforts, the emergent divergence in village development pathways underscores a pressing need for context-specific, classified interventions. To furnish a scientifically grounded typology of villages and inform differentiated development planning, this investigation focuses on Hubei Province as an illustrative case. Synthesizing survey data from 32,457 villages, we developed a multidimensional evaluation framework encompassing four pivotal domains: economic vitality, social service provision, ecological integrity, and cultural value. Leveraging the Self-Organizing Feature Map (SOFM) neural network—an unsupervised machine learning algorithm—we performed a cluster analysis on multi-source, heterogeneous datasets. This technique enabled the objective delineation of spatial typological patterns among Hubei’s villages, elucidated their underlying classification architecture shaped by multifaceted drivers, and demonstrated the methodological robustness and applicability of this approach for large-scale village categorization. Grounded in the derived typologies and informed by strategic directives from higher-tier planning instruments, we conducted a nuanced examination of the distinctive attributes characterizing each village type. The findings provide scientific evidence and decision-making support for village classification and rural revitalization planning in Hubei Province, with valuable implications for other regions with similar development foundations in China. Full article

Background: Imaging-transcriptomic analysis, through the integration of multimodal magnetic resonance imaging (MRI) and transcriptomic data, provides complementary structural, functional, and molecular information that is crucial for the early detection and mechanistic exploration of Alzheimer’s disease (AD). However, effectively extracting features from heterogeneous multimodal data and capturing the associations between microscopic molecular variations and macroscopic brain alterations remain key challenges. Recent advances in deep learning and multimodal integration have enhanced the ability to model nonlinear cross-modal relationships, enabling more accurate identification of imaging-transcriptomic biomarkers and subtypes. Developing robust multimodal frameworks is therefore essential for early AD detection, subtype identification, and advancing precision medicine in neurodegenerative diseases. Methods: In this study, a two-stage method of multimodal Feature Extraction based on Association Analysis and Graph Convolutional Network with Self-Attention and Self-Expression framework (MFEAA-GCNSASE) for early diagnosis of AD and effective identification of subtypes of MCI with different progression to AD is proposed. In the first stage, the MFEAA model is applied to integrate multiple association analysis methods on sMRI, PET, and transcriptomic data to identify key multimodal biomarkers for AD and mild cognitive impairment (MCI). In the second stage, the GCNSASE model enhances classification accuracy between AD and MCI patients through self-attention and self-expression layers. Additionally, unsupervised clustering was performed on MCI samples using top multimodal biomarkers to explore subtype heterogeneity and conversion risk. Reliable MCI subtypes were also identified through a consensus clustering approach. Results: The proposed algorithm integrates sMRI, PET, and transcriptomic data, identifying robust biomarkers including the Left Hippocampus, Left Angular Gyrus, and key genes such as SLC25A5 and GABARAP. To ensure statistical robustness given the extreme class imbalance, we employed a rigorous repeated stratified cross-validation (RSCV) framework. GCNSASE achieved state-of-the-art discrimination performance with mean AUC values ranging from 0.946 to 0.961 across feature subsets (10–50%), significantly outperforming MOGONET (mean AUC: 0.844–0.875, p < 0.001) and conventional machine learning models with tighter 95% confidence intervals, indicating superior stability despite the limited AD sample size. Clustering analysis revealed two distinct MCI subtypes with divergent molecular landscapes: Subtype A was enriched in energy metabolism and cellular maintenance pathways, whereas Subtype B was enriched in neuroinflammatory and aberrant signaling pathways. Notably, the majority of MCI patients who subsequently converted to AD were concentrated in the immune-inflammatory Subtype B. These findings highlight that neuroinflammation coupled with bioenergetic failure constitutes a critical mechanism driving the conversion from MCI to AD. Conclusions: The proposed methods not only provide the key multimodal biomarkers and enhance the accuracy of the classification model for early AD diagnosis but also identify biologically and clinically meaningful MCI subtypes with distinct molecular signatures and conversion risks. Exploring these associated multimodal biomarkers and MCI subtypes is of great significance, as they help elucidate the heterogeneous mechanisms underlying AD onset and progression, enable the identification of high-risk individuals likely to convert to AD, and provide a foundation for targeted therapeutic strategies and individualized clinical management. These findings have important implications for understanding disease heterogeneity, discovering potential intervention targets, and advancing precision medicine in neurodegenerative diseases. Full article

The electric vehicles (EVs) market keeps growing globally; thus, it is critical to secure the EV charging communication protocols in order to guarantee reliable and fair charging operations among the customers. The Open Charge Point Protocol (OCPP) 2.0.1 supports the communication between the Electric Vehicle Supply Equipment (EVSE) and Charging Station Management Systems (CSMSs); therefore, it becomes vulnerable to several types of attacks, which aim to jeopardize smart charging, billing, and energy management. Specifically, OCPP 2.0.1 allows the self-reporting of the State of Charge (SOC) values, which makes it vulnerable to spoofing-based cyberattacks, which target manipulating the scheduling priorities, distorting the load forecasts, and extending the charging sessions in an unfair manner. In this paper, we try to address this type of attack by providing a comprehensive analysis of the SOC spoofing attacks and introducing a novel unsupervised detection framework based on the One-Class Support Vector Machine (OCSVM) algorithm. Specifically, two types of attack scenarios are analyzed (i.e., priority manipulation and session extension) by deriving engineered features that capture the nonlinear relationships under normal charging behavior. Detailed simulation-based results are derived by utilizing the DESL-EPFL Level 3 EV charging dataset. Our results demonstrate high F1-score and recall in identifying spoofed SOC values and that the proposed OCSVM model demonstrates superior performance compared to alternative clustering and deep-learning based detectors. Full article

Existing 3D reconstruction algorithms commonly struggle with modeling specific local objects within large-scale scenes, often resulting in a lack of local detail and incomplete geometric structures. While current mainstream point cloud completion methods can restore these missing structures to some degree, they are fundamentally based on generative in-filling, a process that relies on geometric priors learned from large-scale datasets. Consequently, the physical realism and geometric accuracy of the results cannot be guaranteed. To address these limitations, this paper proposes a novel, data-driven framework for point cloud completion. Our core method involves the high-precision, heterogeneous data registration and seamless fusion of an object-level point cloud—reconstructed with high-fidelity appearance and geometry by our optimized Neural Radiance Fields (NeRF) framework—with our target large-scale scene point cloud. By using high-precision, physically based data as a strong prior for geometric completion, we offer an alternative route to conventional generative completion methods. Concurrently, we employ unsupervised evaluation metrics to assess the intrinsic quality of the final results. This work provides a robust and high-fidelity solution to the problem of completing local objects within large-scale scenes. Evaluated on our self-constructed UAV-Recon dataset, the proposed method achieved a Structural Plausibility ≥ 0.995, Geometric Smoothness ≤ 0.19, and Distribution Uniformity ≈ 1.2, offering a robust solution for the high-fidelity completion of local objects within large-scale scenes. Full article

In the financial sector, machine learning has become essential for credit risk assessment, often outperforming traditional statistical approaches, such as linear regression, discriminant analysis, or model-based expert judgment. Although machine learning technologies are increasingly being used, further research is needed to understand how they can be effectively combined and how different models interact during credit evaluation. This study proposes a technique that integrates hierarchical clustering, namely Agglomerative clustering and Balanced Iterative Reducing and Clustering using Hierarchies, along with individual supervised models and a self organizing map-based consensus model. This approach helps to better understand how different clustering algorithms influence model performance. To support this approach, we performed a detailed unsupervised component analysis using metrics such as the silhouette score and Adjusted Rand Index to assess cluster quality and its relationship with the classification results. The study was applied to multiple datasets, including a Taiwanese credit dataset. It was also extended to a multiclass classification scenario to evaluate its generalization ability. The results show that the quality metrics of the cluster correlate with the performance, highlighting the importance of combining unsupervised clustering and self organizing map consensus methods for improving credit evaluation. Full article

The problem of source-code analysis using machine-learning techniques has gained much attention recently, as several powerful code-embedding methods have been created. Having different embedding methods available for source code has opened the way to tackling many practical problems in source-code analysis. This paper addresses the problem of determining the number of distinct algorithmic strategies that may be found in a set of correct solutions to a competitive programming problem. To achieve this, we employ a novel unsupervised algorithm that uses a multiview interpretation of data based on different embedding and clustering methods, a multidimensional assignment problem (MAP) to determine a subset of a higher probability of correctness, and a self-training method based on voting to determine the correct clusters of the remaining set. We investigate the following two aspects: (1) whether the proposed unsupervised approach outperforms existing methods when the number K of distinct algorithmic strategies is known and (2) Whether the approach can also be applied to determine the optimal value of K. We have addressed these using seven embedding methods with three clustering strategies in a data-analysis pipeline that tackles the previously described issues on a newly created dataset consisting of 15 algorithmic problems. According to the results, for the first aspect, the proposed unsupervised voting algorithm significantly improves the baseline clustering approach for a known K. This improvement was observed across all problems in the dataset, except one. In the case of the second one, we prove that the proposed method has a negative impact on determining the optimal number K. Scale-up of the data-analysis pipeline to datasets of thousands of problems may yield the ability to profoundly understand and learn about the innovative process of correctly designing and writing code in the context of competitive programming or even industry code. Full article

Object detection in degraded environments such as low-light and nighttime conditions remains a challenging task, as conventional computer vision techniques often fail to achieve high precision and robust performance. With the increasing adoption of deep learning, this paper aims to enhance object detection under such adverse conditions by proposing an improved version of YOLOv11, named DRF-YOLO (Degradation-Robust and Feature-enhanced YOLO). The proposed framework incorporates three innovative components: (1) a lightweight Cross Stage Partial Multi-Scale Edge Enhancement (CSP-MSEE) module that combines multi-scale feature extraction with edge enhancement to strengthen feature representation; (2) a Focal Modulation attention mechanism that improves the network’s responsiveness to target regions and contextual information; and (3) a self-developed Dynamic Interaction Head (DIH) that enhances detection accuracy and spatial adaptability for small objects. In addition, a lightweight unsupervised image enhancement algorithm, Zero-DCE (Zero-Reference Deep Curve Estimation), is introduced prior to training to improve image contrast and detail, and Generalized Intersection over Union (GIoU) is employed as the bounding box regression loss. To evaluate the effectiveness of DRF-YOLO, experiments are conducted on two representative low-light datasets: ExDark and the nighttime subset of BDD100K, which include images of vehicles, pedestrians, and other road objects. Results show that DRF-YOLO achieves improvements of 3.4% and 2.3% in mAP@0.5 compared with the original YOLOv11, demonstrating enhanced robustness and accuracy in degraded environments while maintaining lightweight efficiency. Full article

Recent advancements in composite materials design have made glass fiber-reinforced polymer composites (GFRPC) a viable choice for a wide range of engineering and industrial applications. Although GFRPCs boast attractive characteristics such as low specific mass and high specific mechanical strength, identifying and characterizing damage mechanisms in these materials is challenging. Several scientific studies have examined the root causes of GFRPC failure using various methods, including non-destructive techniques and learning algorithms. Despite this, ongoing investigations aim to accurately detect mechanical defects in GFRPCs. This study explores the use of non-destructive testing (NDT) combined with unsupervised learning algorithms to identify and classify damage mechanisms in GFRPCs. The NDT method employed in this study is acoustic emission (AE), which identifies waveforms associated with various failure mechanisms during testing. These waveforms are categorized using unsupervised learning methods such as principal component analysis (PCA) and self-organizing maps. PCA selects the most appropriate AE descriptors for distinguishing between different damage mechanisms, while the self-organizing maps algorithm performs clustering analysis and classifies failure mechanisms. Scanning electron microscope images of the observed failures are provided to sup-port the findings derived from AE data. Full article

The construction and maintenance of an ecological security pattern (ESP) are important for promoting the regional development of ecological civilizations, realizing sustainable and healthy development, and creating a harmonious and beautiful space for human beings and nature to thrive. Traditional construction methods have the limitations of a single dimension, a single method, and excessive human subjective intervention for source and corridor identification, without considering the multidimensional quality of the sources and the structural connectivity and resilience optimization of the corridors. Therefore, an ecological civilization demonstration area (Jiangxi Province) was used as the study area, a new research method for ESP was proposed, and an empirical study was conducted. To evaluate ecosystem service (ES) importance–disturbance–risk and extract sustainability sources through the deep embedded clustering–self-organizing map (DEC–SOM) deep unsupervised learning clustering algorithm, ecological networks (ENs) were constructed by applying the minimum cumulative resistance (MCR) gravity model and circuit theory. The ENs were then optimized to improve performance by combining the comparative advantages of the two approaches in terms of structural connectivity and resilience. A comparative analysis of EN performance was constructed among different functional control zones, and the ESP was constructed to include 42 ecological sources, 134 corridors, 210 restoration nodes, and 280 protection nodes. An ESP of ‘1 nucleus, 3 belts, 6 zones, and multiple corridors’ was constructed, and the key restoration components and protection functions were clarified. This study offers a valuable reference for ecological management, protection, and restoration and provides insights into the promotion of harmonious symbiosis between human beings and nature and sustainable regional development. Full article

We propose a tri-fold learning algorithm, called Neuroevolution of Hybrid Neural Networks in a Robotic Agent (acronym in Spanish, NRNH-AR), based on deep reinforcement learning (DRL), with self-supervised learning (SSL) and unsupervised learning (USL) steps, specifically designed to be implemented in a small autonomous navigation robot capable of operating in constrained physical environments. The NRNH-AR algorithm is designed for a small physical robotic agent with limited resources. The proposed algorithm was evaluated in four critical aspects: computational cost, learning stability, required memory size, and operation speed. The results obtained show that the performance of NRNH-AR is within the ranges of the Deep Q Network (DQN), Deep Deterministic Policy Gradient (DDPG), and Twin Delayed Deep Deterministic Policy Gradient (TD3). The proposed algorithm comprises three types of learning algorithms: SSL, USL, and DRL. Thanks to the series of learning algorithms, the proposed algorithm optimizes the use of resources and demonstrates adaptability in dynamic environments, a crucial aspect of navigation robotics. By integrating computer vision techniques based on a Convolutional Neuronal Network (CNN), the algorithm enhances its abilities to understand visual observations of the environment rapidly and detect a specific object, avoiding obstacles. Full article

Canada’s vast forests play a substantial role in the global carbon balance but require laborious and expensive forest inventory campaigns to monitor changes in aboveground biomass (AGB). Light detection and ranging (LiDAR) or reflectance observations onboard airborne or unoccupied aerial vehicles (UAVs) may address scalability limitations associated with traditional forest inventory but require simple forest structures or large sets of manually delineated crowns. Here, we introduce a deep learning approach for crown delineation and AGB estimation reproducible for complex forest structures without relying on hand annotations for training. Firstly, we detect treetops and delineate crowns with a LiDAR point cloud using marker-controlled watershed segmentation (MCWS). Then we train a deep learning model on annotations derived from MCWS to make crown predictions on UAV red, blue, and green (RGB) tiles. Finally, we estimate AGB metrics from tree height- and crown diameter-based allometric equations, all derived from UAV data. We validate our approach using 14 ha mixed forest stands with various experimental tree densities in Southern Ontario, Canada. Our results show that using an unsupervised LiDAR-only algorithm for tree crown delineation alongside a self-supervised RGB deep learning model trained on LiDAR-derived annotations leads to an 18% improvement in AGB estimation accuracy. In unharvested stands, the self-supervised RGB model performs well for height (adjusted R², R_a² = 0.79) and AGB (R_a² = 0.80) estimation. In thinned stands, the performance of both unsupervised and self-supervised methods varied with stand density, crown clumping, canopy height variation, and species diversity. These findings suggest that MCWS can be supplemented with self-supervised deep learning to directly estimate biomass components in complex forest structures as well as atypical forest conditions where stand density and spatial patterns are manipulated. Full article

Specific emitter identification (SEI), as an emerging physical-layer security authentication method, is crucial for maintaining information security in the Internet of Things. However, existing deep learning-based SEI methods require extensive labeled data for training, which are often unavailable in untrusted scenarios. Furthermore, due to the subtle nature of radio-frequency fingerprints, unsupervised SEI struggles to achieve high accuracy in identification without the guidance of labels. In this paper, we propose an unsupervised SEI method based on group label-driven contrastive learning (GLD-CL). We propose a novel method for constructing the dataset: all input samples derived from the same received signal segment are grouped together and assigned a unique identifier, termed the group label. Based on this, we improve the loss function of self-supervised contrastive learning. With the assistance of group labels, the feature vectors of the same class in the feature space become more closely clustered, enhancing the accuracy of unsupervised SEI. Extensive experimental results based on real-world datasets demonstrate that the normalized mutual information of GLD-CL achieves 96.4% accuracy, representing an improvement of 5.68% or more compared to the baseline algorithms. Furthermore, GLD-CL exhibits robust performance, achieving good identification accuracy across various signal-to-noise ratio scenarios. Full article

Left ventricular ejection fraction (LVEF) is a critical indicator for assessing cardiac function and diagnosing heart disease. LVEF can be derived by estimating the left ventricular volume from end-systolic and end-diastolic frames through echocardiography segmentation. However, current algorithms either focus primarily on single-frame segmentation, neglecting the temporal and spatial correlations between consecutive frames, or often fail to effectively address the inherent challenges posed by the low-contrast and fuzzy edges characteristic of echocardiography, thereby resulting in suboptimal segmentation outcomes. In this study, we propose a novel self-attention-based dual-task learning approach for automatic left ventricle segmentation. First, we introduce a multi-scale edge-attention U-Net to achieve supervised semantic segmentation of echocardiography. Second, an optical flow network is developed to capture the changes in the optical flow fields between frames in an unsupervised manner. These two tasks are then jointly trained using a temporal consistency mechanism to extract spatio-temporal features across frames. Experimental results demonstrate that our model outperforms existing segmentation methods. Our proposed method not only enhances the performance of semantic segmentation but also improves the consistency of segmentation between consecutive frames. Full article

Histopathological images stained with hematoxylin and eosin (H&E) are crucial for cancer diagnosis and prognosis. However, color variations caused by differences in tissue preparation and scanning devices can lead to data distribution discrepancies, adversely affecting the performance of downstream algorithms in tasks like classification, segmentation, and detection. To address these issues, stain normalization methods have been developed to standardize color distributions across images from various sources. Recent advancements in deep learning-based stain normalization methods have shown significant promise due to their minimal preprocessing requirements, independence from reference templates, and robustness. This review examines 115 publications to explore the latest developments in this field. We first outline the evaluation metrics and publicly available datasets used for assessing stain normalization methods. Next, we systematically review deep learning-based approaches, including supervised, unsupervised, and self-supervised methods, categorizing them by core technologies and analyzing their contributions and limitations. Finally, we discuss current challenges and future directions, aiming to provide researchers with a comprehensive understanding of the field, promote further development, and accelerate the progress of intelligent cancer diagnosis. Full article