Search Results (29)

Multimodal depression detection aims to identify depressive states by integrating complementary affective cues from audio, visual, and textual modalities. However, existing approaches often directly fuse multimodal features, which may introduce semantic interference, information redundancy, and insufficiently discriminative representations. To address these issues, this paper proposes a multimodal depression detection model based on modality feature decomposition and contrastive learning (MFDCL). Specifically, unimodal representations are decomposed into modality-shared and modality-specific features to distinguish cross-modal affective consistency from modality-dependent characteristics. Subsequently, sample-to-sample contrastive learning is applied within the shared feature space, where label-guided positive and negative sample pairs are constructed to enhance the discriminability between depressed and non-depressed states. In addition, a lightweight multi-scale attention fusion module is introduced to capture affective cues across different receptive fields. Experimental results on the DAIC-WOZ and E-DAIC datasets demonstrate the effectiveness of the proposed method. On DAIC-WOZ, MFDCL achieves Precision, Recall, F1-score, and Accuracy of 0.92, 0.90, 0.91, and 0.94, respectively, while on E-DAIC, it achieves 0.95, 0.85, 0.90, and 0.97. These results indicate that MFDCL effectively enhances feature discriminability and achieves competitive performance in multimodal depression detection. Full article

Extracellular vesicle (EV) transcriptomic data provides a high-dimensional representation of cellular states but remains challenging to interpret due to noise, redundancy, and limited sample sizes. Most existing approaches rely on supervised differential expression analyses, which can be biased and may fail to capture latent structure in small datasets. In this study, we propose an unsupervised machine learning framework based on non-negative matrix factorization (NMF) to identify latent expression programs from EV mRNA profiles. A structured preprocessing pipeline combining expression filtering, variance selection, ANOVA-based feature selection, and correlation pruning was used to reduce dimensionality and improve signal quality prior to matrix factorization. NMF was applied to decompose the data into interpretable gene modules and sample-specific activation patterns. Model selection was performed using reconstruction error and component stability across multiple initializations. Candidate features were prioritized using a composite ranking score integrating module loadings, group-level expression differences, and model stability. The approach identified a stable low-rank representation capturing dominant patterns in the data and a compact set of informative features. These results demonstrate that unsupervised matrix factorization can effectively extract structured, interpretable signals from small-scale transcriptomic datasets and provide a robust framework for feature prioritization and representation learning in high-dimensional biological data. Full article

Advanced sensor signal analysis is increasingly important for intelligent health management in human-centered environments, where continuous perception and real-time interpretation of motion-related signals are essential for safe and adaptive assistance. In this study, we propose a vision-based sensor signal analysis framework for automated ergonomic risk assessment of wheelchair users during cabinet interaction. The proposed framework integrates YOLOv11 for human detection, MHFormer for monocular 3D pose reconstruction, and a fuzzy logic-enhanced RULA model for continuous ergonomic risk quantification from video-derived motion signals. To support model development and evaluation, we constructed a dedicated wheelchair cabinet-operation dataset comprising 30 participants, including 14 experienced wheelchair users and 16 trained simulation participants, across five representative cabinet-operation scenarios. The raw dataset contained approximately 5 h of RGB video and about 150,000 original frames. To reduce redundancy caused by highly similar consecutive frames and to mitigate overfitting risk, representative frames were sampled from the continuous video sequences, resulting in 10,000 images for annotation and model development. Based on the proposed framework, raw visual sensor signals are transformed into temporally continuous kinematic representations and ergonomic risk scores, enabling non-contact and real-time health-state interpretation in assistive living environments. The proposed method achieved an average joint-angle estimation RMSE of 7.5°, representing an approximately 60% reduction compared with a Kinect v2-based motion capture baseline (18.6°), which is widely used for low-cost ergonomic evaluation. In benchmark evaluation, the proposed method achieved 84% risk-classification accuracy with a Cohen’s kappa of 0.66, outperforming representative baseline approaches. The results further indicated that low revolving-door and low-drawer operations were associated with higher and more sustained ergonomic risk exposure than sliding-door interaction. These findings demonstrate that vision-based sensor signal analysis can provide an effective solution for intelligent health management, ergonomic monitoring, and perception-driven assessment in accessible and assistive autonomous living systems. Full article

Hyperspectral image (HSI) classification is fundamental to a wide range of remote sensing applications, such as precision agriculture, environmental monitoring, and urban planning, because HSIs provide rich spectral signatures that enable the discrimination of subtle material differences. Deep learning approaches, including Convolutional Neural Networks (CNNs), Graph Convolutional Networks (GCNs), and Transformers, have achieved strong performance in learning spatial–spectral representations. However, these models often face difficulties in jointly modeling long-range dependencies, fine-grained local structures, and non-Euclidean spatial relationships, particularly when labeled training data are scarce. This paper proposes a Spectral–Spatial Graph Transformer Network (SSGTN), a dual-branch architecture that integrates superpixel-based graph modeling with Transformer-based global reasoning. SSGTN consists of four key components, namely (1) an LDA-SLIC superpixel graph construction module that preserves discriminative spectral–spatial structures while reducing computational complexity, (2) a lightweight spectral denoising module based on $1\times 1$ convolutions and batch normalization to suppress redundant and noisy bands, (3) a Spectral–Spatial Shift Module (SSSM) that enables efficient multi-scale feature fusion through channel-wise and spatial-wise shift operations, and (4) a dual-branch GCN-Transformer block that jointly models local graph topology and global spectral–spatial dependencies. Extensive experiments on three public HSI datasets (Indian Pines, WHU-Hi-LongKou, and Houston2018) under limited supervision (1% training samples) demonstrate that SSGTN consistently outperforms state-of-the-art CNN-, Transformer-, Mamba-, and GCN-based methods in overall accuracy, Average Accuracy, and the $\kappa$ coefficient. The proposed framework provides an effective baseline for HSI classification under limited supervision and highlights the benefits of integrating graph-based structural priors with global contextual modeling. Full article

In hyperspectral image classification (HSIC), each pixel contains information across hundreds of contiguous spectral bands; therefore, the ability to perform long-distance modeling that stably captures and propagates these long-distance dependencies is critical. A selective structured state space model (SSM) named Mamba has shown strong capabilities for capturing cross-band long-distance dependencies and exhibits advantages in long-distance modeling. However, the inherently high spectral dimensionality, information redundancy, and spatial heterogeneity of hyperspectral images (HSI) pose challenges for Mamba in fully extracting spatial–spectral features and in maintaining computational efficiency. To address these issues, we propose S²GL-MambaResNet, a lightweight HSI classification network that tightly couples Mamba with progressive residuals to enable richer global, local, and multi-scale spatial–spectral feature extraction, thereby mitigating the negative effects of high dimensionality, redundancy, and spatial heterogeneity on long-distance modeling. To avoid fragmentation of spatial–spectral information caused by serialization and to enhance local discriminability, we design a preprocessing method applied to the features before they are input to Mamba, termed the Spatial–Spectral Gated Attention Aggregator (SS-GAA). SS-GAA uses spatial–spectral adaptive gated fusion to preserve and strengthen the continuity of the central pixel’s neighborhood and its local spatial–spectral representation. To compensate for a single global sequence network’s tendency to overlook local structures, we introduce a novel Mamba variant called the Global_Local Spatial_Spectral Mamba Encoder (GLS²ME). GLS²ME comprises a pixel-level global branch and a non-overlapping sliding-window local branch for modeling long-distance dependencies and patch-level spatial–spectral relations, respectively, jointly improving generalization stability under limited sample regimes. To ensure that spatial details and boundary integrity are maintained while capturing spectral patterns at multiple scales, we propose a multi-scale Mamba encoding scheme, the Hierarchical Spectral Mamba Encoder (HSME). HSME first extracts spectral responses via multi-scale 1D spectral convolutions, then groups spectral bands and feeds these groups into Mamba encoders to capture spectral pattern information at different scales. Finally, we design a Progressive Residual Fusion Block (PRFB) that integrates 3D residual recalibration units with Efficient Channel Attention (ECA) to fuse multi-kernel outputs within a global context. This enables ordered fusion of local multi-scale features under a global semantic context, improving information utilization efficiency while keeping computational overhead under control. Comparative experiments on four publicly available HSI datasets demonstrate that S²GL-MambaResNet achieves superior classification accuracy compared with several state-of-the-art methods, with particularly pronounced advantages under few-shot and class-imbalanced conditions. Full article

Digital surface models (DSMs) derived from high-resolution satellite imagery often contain mismatches, voids, and coarse building geometry, limiting their suitability for accurate and standardized 3D reconstruction. The scarcity of finely annotated samples further constrains generalization to complex structures. To address these challenges, an automated building reconstruction method based on two-stage polygon decomposition and adaptive roof fitting is proposed. Building polygons are first extracted and standardized to preserve primary contours while improving geometric regularity. A two-stage decomposition is then applied. In the first stage, polygons are coarsely decomposed, and redundant rectangles are removed by analyzing containment relationships. In the second stage, non-flat regions are identified and further decomposed to accommodate complex building connections. For 3D model fitting, flat-roof buildings are reconstructed by integrating structural analysis of DSM elevation distributions with adaptive rooftop partitioning, which enables accurate modeling of complex flat structures with auxiliary components. For non-flat roofs, a representative parameter space is defined and explored through systematic search and optimization to obtain precise fits. Finally, intersecting primitives are normalized and optimally merged to ensure structural coherence and standardized representation. Experiments on the US3D, MVS3D, and Beijing-3 datasets demonstrate that the proposed method achieves higher geometric accuracy and more standardized models, with an average IOU3 of 91.26%, RMSE of 0.78 m, and MHE of 0.22 m. Full article

In this article, a novel procedure is developed to properly handle the 3-D mispositioning of the scanning probe in the near-field to far-field (NFtFF) transformations with spherical scanning for quasi-planar antennas under test, which make use of a non-redundant (NR) number of samples. It proceeds through two stages. In the former, a phase correction technique, named spherical wave correction, is applied to compensate for the phase shifts of the collected NF samples, which do not belong to the measurement sphere, due to mechanical defects of the arc, or inaccuracy of the robotic arm employed in the considered NF facility driving the probe. Once the phase shifts have been compensated, the recovered NF samples belong to the set spherical surface, but their positions differ from those prescribed by the adopted NR representation, because of an imprecise control and/or inaccuracy of the positioning system. Thus, the resulting sampling arrangement is affected by 2-D mispositioning errors. Accordingly, an iterative procedure is used in the latter step to restore the NF samples at their exact locations from those determined at the first step. Once the correct sampling arrangement has been retrieved from the 3-D mispositioned one, an optimal sampling interpolation formula is employed to obtain the massive input NF data necessary for the classical spherical NFtFF transformation technique. Numerical results, showing the precision of the NF and FF reconstructions, assessed the efficacy of the developed procedure. Full article

Tall fescue toxicosis, caused by ingestion of endophyte-infected tall fescue (Lolium arundinaceum), impairs growth and reproduction in beef cattle and results in over USD 3 billion annual loss to the U.S. livestock industry. While the effects on host metabolism and rumen function have been described, the impact on the rectal microbiome remains poorly understood. In this study, we performed whole-genome shotgun metagenomic sequencing on fecal samples collected before and after a 30-day toxic fescue seed supplementation from eight pregnant Angus × Simmental cows and heifers. We generated 157 Gbp of sequencing data in 16 metagenomes, and assembled 13.1 Gbp de novo microbial contigs, identifying 22 million non-redundant microbial genes from the cattle rectum microbiome. Fescue toxicosis significantly reduced alpha diversity (p < 0.01) and altered beta diversity (PERMANOVA p < 0.01), indicating microbial dysbiosis. We discovered significant enrichment of 31 bacterial species post-treatment, including multiple core rumen taxa. Ruminococcaceae bacterium P7 showed an average of 16-fold increase in fecal abundance (p < 0.01), making it the top-featured species in linear discriminant analysis. Functional pathway analysis revealed a shift from energy metabolism to antimicrobial resistance and DNA replication following toxic seed consumption. Comparative analysis showed increased representation of core rumen taxa in rectal microbiota post-treatment, suggesting disrupted rumen function. These findings demonstrate that fescue toxicosis alters both the composition and functional landscape of the hindgut microbiota. Ruminococcaceae bacterium P7 emerges as a promising biomarker for monitoring fescue toxicosis through non-invasive fecal sampling, with potential applications in herd-level diagnostics and mitigation strategies. Full article

In this study, we propose a novel pixel-level non-local similarity (PNS)-based reconstruction method for millimeter-wave interferometric synthetic aperture radiometer (InSAR) imaging. Unlike traditional compressed sensing (CS) methods, which rely on predefined sparse transforms and often introduce artifacts, our approach leverages structural redundancies in InSAR images through an enhanced sparse representation model with dynamically filtered coefficients. This design simultaneously preserves fine details and suppresses noise interference. Furthermore, an iterative refinement mechanism incorporates raw sampled data fidelity constraints, enhancing reconstruction accuracy. Simulation and physical experiments demonstrate that the proposed InSAR-PNS method significantly outperforms conventional techniques: it achieves a 1.93 dB average peak signal-to-noise ratio (PSNR) improvement over CS-based reconstruction while operating at reduced sampling ratios compared to Nyquist-rate fast fourier transform (FFT) methods. The framework provides a practical and efficient solution for high-fidelity millimeter-wave InSAR imaging under sub-Nyquist sampling conditions. Full article

This study aimed to determine the practical efficacy of passive eDNA samplers (PEDS) for monitoring fish diversity in riverine ecosystems. It investigated the utility of environmental DNA (eDNA) in accurately depicting fish composition and diversity within the Lancang River. Environmental DNA technology, particularly PEDS, may be used as a substitute for traditional water filtration techniques. However, its effectiveness in natural water ecosystems remains to be proven. The filter materials included mixed cellulose acetate and nitrate (MCE), nylon (NL), glass fiber (GF), and polyvinyl chloride filter membrane (PVC). This study used four different types of filters, each with identical pore sizes and dimensions but constructed from various materials, to assess eDNA capture under laboratory and field conditions in the water samples. The filter materials included mixed cellulose acetate and nitrate (MCE), nylon (NL), glass fiber (GF), and polyvinyl chloride filter membrane (PVC). Environmental DNA macrobarcoding was used to analyze fish biodiversity and to understand the environmental effects on species distribution. Our study identified 50 fish species inhabiting the Lancang River, with equal representation of exotic and native species. A comparative analysis of four filter-based environmental DNA samplers and traditional environmental DNA sampling methods demonstrated comparable species richness. Redundancy analysis indicated that environmental variables, elevation, electrical conductivity, salinity, and chlorophyll-a significantly influenced the distribution patterns of both non-native and native fish species in the river. This study highlights the significance of eDNA technology in evaluating fish diversity across diverse habitats, thereby establishing a theoretical framework for the sustained monitoring and management of fish biodiversity in protected areas. Full article

Single image super-resolution (SISR) aims to reconstruct high-quality images from low-resolution inputs, a persistent challenge in computer vision with critical applications in medical imaging, satellite imagery, and video enhancement. Traditional diffusion model-based (DM-based) methods, while effective in restoring fine details, suffer from computational inefficiency due to their iterative denoising process. To address this, we introduce the Sparse Momentum-based Faster Diffusion Model (SMFDM), designed for rapid and high-fidelity super-resolution. SMFDM integrates a novel encoder state reuse mechanism that selectively omits non-critical time steps during the denoising phase, significantly reducing computational redundancy. Additionally, the model employs a sparse momentum mechanism, enabling robust representation capabilities while utilizing only a fraction of the original model weights. Experiments demonstrate that SMFDM achieves an impressive 71.04% acceleration in the diffusion process, requiring only 15% of the original weights, while maintaining high-quality outputs with effective preservation of image details and textures. Our work highlights the potential of combining sparse learning and efficient sampling strategies to enhance the practical applicability of diffusion models for super-resolution tasks. Full article

This paper concerns the experimental validation of optimized near-field (NF) spherical spiral scannings employing a minimum number of samples, when an offset-mounted elongated antenna under test (AUT), i.e., with its center shifted with respect to that of the measurement sphere, is considered. In order to perform the standard NF/far-field transformation (NF/FFT) technique, a non-centered AUT would generally require the collection of a significantly increased amount of voltage data if compared to the onset scenario. This issue is addressed here by using the non-redundant (NR) sampling representations of electromagnetic (EM) fields. These representations, by leveraging the analytical properties of the EM field, allow one to perform the NR NF/FFTs for offset-mounted AUTs by using only a minimum number of (offset acquired) samples, equal to that required by the NR approaches for the onset case (over 85% fewer samples compared to the standard NF spherical scanning). In particular, these NR NF/FFTs are obtained by modeling the AUT with a prolate spheroid or a rounded cylinder and their effectiveness is fully assessed by the reported experimental results. Full article

Existing RGB image-based object detection methods achieve high accuracy when objects are static or in quasi-static conditions but demonstrate degraded performance with fast-moving objects due to motion blur artifacts. Moreover, state-of-the-art deep learning methods, which rely on RGB images as input, necessitate training and inference on high-performance graphics cards. These cards are not only bulky and power-hungry but also challenging to deploy on compact robotic platforms. Fortunately, the emergence of event cameras, inspired by biological vision, provides a promising solution to these limitations. These cameras offer low latency, minimal motion blur, and non-redundant outputs, making them well suited for dynamic obstacle detection. Building on these advantages, a novel methodology was developed through the fusion of events with depth to address the challenge of dynamic object detection. Initially, an adaptive temporal sampling window was implemented to selectively acquire event data and supplementary information, contingent upon the presence of objects within the visual field. Subsequently, a warping transformation was applied to the event data, effectively eliminating artifacts induced by ego-motion while preserving signals originating from moving objects. Following this preprocessing stage, the transformed event data were converted into an event queue representation, upon which denoising operations were performed. Ultimately, object detection was achieved through the application of image moment analysis to the processed event queue representation. The experimental results show that, compared with the current state-of-the-art methods, the proposed method has improved the detection speed by approximately 20% and the accuracy by approximately 5%. To substantiate real-world applicability, the authors implemented a complete obstacle avoidance pipeline, integrating our detector with planning modules and successfully deploying it on a custom-built quadrotor platform. Field tests confirm reliable avoidance of an obstacle approaching at approximately 8 m/s, thereby validating practical deployment potential. Full article

A feasibility study on the compensation of 3D mispositioning errors of the probe occurring in the characterization of a long antenna, via a non-redundant (NR) near to far-field (NTFF) transformation with helicoidal scan, is conducted in this article. Such types of errors can result from imperfections in the rail driving the linear motion of the probe and from an imprecise synchronization of the linear and rotational movements of the probe and the antenna when drawing the scan helix. To correct them, an approach, which proceeds through two steps, is proposed. The former step uses a technique called cylindo rical wave (CW) correction for compensating the phase of the near-field (NF) samples, which, owing to the rail imperfections, result in not being acquired over the measurement cylinder surface. The latter exploits an iterative scheme to restore the samples at the sampling points required by the adopted NR representation along the scan helix from those obtained by applying the CW correction technique and impaired by 2D mispositioning errors. The so compensated NF samples are then effectively recovered via a 2D optimal sampling interpolation (OSI) scheme to accurately obtain the input data required to carry out the standard cylindrical NTFF transformation. The OSI representation is determined here by assuming a long antenna under test as enclosed in a prolate ellipsoid or cylinder ending into two hemispheres (cigar) in order to make, depending on the particular geometry of the considered antenna, the representation effectively non-redundant. The reported numerical simulation results show the capability of the proposed approach to compensate even severe 3D mispositioning errors, thus enabling its usage in a real measurement scenario. Full article

This paper deals with the compensation of the probe mispositioning errors occurring in a cylindrical near-field (NF) facility due to the imprecise control of the linear and azimuthal positioners allowing the cylindrical scanning and/or to their limited resolution and to defects in the rails guiding the linear motion. As a result, 3-D errors in the positioning of the probe at any sampling point, as prescribed by the adopted non-redundant representation, affect the accuracy of the NF measurements. An efficient procedure is here proposed to properly compensate for these errors. It involves two steps. The former allows one to correct the mispositioning errors due to the deviation of each actual sampling point from the nominal measurement cylinder. The latter makes use of an iterative technique to restore the NF samples at any sampling point fixed by the used non-redundant representation from the ones obtained at the previous step and affected by 2-D mispositioning errors. Once these steps have been fruitfully applied, the so-compensated NF samples are effectively interpolated through a 2-D optimal sampling interpolation (OSI) formula to accurately reconstruct the input data required to perform the traditional cylindrical near-to-far-field transformation. The OSI representation is here developed by considering an elongated antenna under test as enclosed either in a prolate spheroid or in a cylinder terminated by two half spheres (rounded cylinder) in order to make the representation effectively non-redundant. Numerical test results, which thoroughly prove the efficacy of the devised procedure in correcting even severe 3-D mispositioning errors, are reported. Full article