Search Results (307)

Real-time object detection in Unmanned Aerial Vehicle (UAV) imagery presents unique challenges, primarily characterized by extreme scale variations and intense background clutter. Existing detectors often suffer from spectral homogenization in which the critical high-frequency details of minute targets are washed out by dominant background signals during feature downsampling. To address this, we propose Eagle-YOLO, a dynamic feature aggregation framework designed to master these complexities without compromising inference speed. We introduce three core innovations: (1) the Hierarchical Granularity Block (HG-Block), which employs a residual granularity injection pathway to function as a detail anchor for tiny objects while simultaneously accumulating semantics for large structures; (2) the Cross-Stage Context Modulation (CSCM) mechanism, which leverages a global context query to filter background redundancy and recalibrate features across network stages; and (3) the Scale-Adaptive Heterogeneous Convolution (SAHC) strategy, which dynamically aligns receptive fields with the inherent scale distribution of aerial data. Extensive experiments on the DUT Anti-UAV dataset demonstrate that Eagle-YOLO achieves a remarkable balance between accuracy and latency. Specifically, our lightweight Eagle-YOLO-T variant achieves 74.62% AP, surpassing the robust baseline RTMDet-T by 1.67% while maintaining a real-time inference speed of 141 FPS on an NVIDIA RTX 4090 GPU. Furthermore, on the challenging Anti-UAV dataset, our Eagle-YOLOv8-M variant reaches an impressive 94.38% ${\mathrm{AP}}_{50}^{\mathrm{val}}$, outperforming the standard YOLOv8-M by 2.83% and proving its efficacy for edge-deployed aerial surveillance applications. Full article

Rice seed variety classification is crucial for seed quality control and breeding, yet practical deployment is often limited by the computational and memory demands of modern deep models. We propose SimpleEfficientCNN (SimpleEfficient: simple & efficient; CNN: convolutional neural network), an ultra-lightweight convolutional network built on depthwise separable convolutions for efficient fine-grained seed classification. Experiments were conducted on three datasets with distinct imaging characteristics: a self-constructed Guangdong dataset (7 varieties; 10,500 seeds imaged once and expanded to 112 K images via post-split augmentation), the public M600 rice subset (7 varieties; 9100 original images expanded to 112 K images using the same post-split augmentation pipeline for scale-matched comparison), and the International dataset (75 K images; official train/validation/test split provided by the original release and used as-is without any preprocessing or augmentation, 5 varieties). SimpleEfficientCNN achieved 98.52%, 88.07%, and 99.37% accuracy on the Guangdong, M600, and International test sets, respectively. With only 0.231 M parameters (≈92× fewer than ResNet34), it required 20.5 MB peak GPU memory and delivered 2.0 ms GPU latency (RTX 4090D, batch = 1, FP32) and 1.8 ms single-thread CPU median latency (Ryzen 9 7950X3D, batch = 1, FP32). These results indicate that competitive accuracy can be achieved with substantially reduced model size and inference cost, supporting deployment in resource-constrained agricultural settings. Full article

Automated anomaly detection in transportation infrastructure is essential for enhancing safety and reducing the operational costs associated with manual inspection protocols. This study presents an improved neuromorphic vision system, which extends the prior SIFT-SNN (scale-invariant feature transform–spiking neural network) proof-of-concept by incorporating temporal feature aggregation for context-aware and sequence-stable detection. Analysis of classical stitching-based pipelines exposed sensitivity to motion and lighting variations, motivating the proposed temporally smoothed neuromorphic design. SIFT keypoints are encoded into latency-based spike trains and classified using a leaky integrate-and-fire (LIF) spiking neural network implemented in PyTorch. Evaluated across three hardware configurations—an NVIDIA RTX 4060 GPU, an Intel i7 CPU, and a simulated Jetson Nano—the system achieved 92.3% accuracy and a macro F1 score of 91.0% under five-fold cross-validation. Inference latencies were measured at 9.5 ms, 26.1 ms, and ~48.3 ms per frame, respectively. Memory footprints were under 290 MB, and power consumption was estimated to be between 5 and 65 W. The classifier distinguishes between safe, partially dislodged, and fully dislodged barrier pins, which are critical failure modes for the Auckland Harbour Bridge’s Movable Concrete Barrier (MCB) system. Temporal smoothing further improves recall for ambiguous cases. By achieving a compact model size (2.9 MB), low-latency inference, and minimal power demands, the proposed framework offers a deployable, interpretable, and energy-efficient alternative to conventional CNN-based inspection tools. Future work will focus on exploring the generalisability and transferability of the work presented, additional input sources, and human–computer interaction paradigms for various deployment infrastructures and advancements. Full article

Background and Objectives: The quality of recipient vessels is critical for successful microsurgical breast reconstruction, and iatrogenic damage should be minimized. Adjuvant radiotherapy (RTx) and chemotherapy (CTx) are widely used for breast cancer and may induce structural changes in recipient vessels. This study aimed to evaluate changes in recipient vessel diameters for breast reconstruction after adjuvant treatment in patients with breast cancer. Materials and Methods: A total of 167 patients with unilateral breast cancer who underwent surgical resection between 2017 and 2021 were retrospectively reviewed. Patients were classified into four groups: mastectomy only without adjuvant treatment (group A, n = 33), adjuvant RTx only (group B, n = 44), adjuvant CTx only (group C, n = 43), and combined adjuvant CTx and RTx (group D, n = 47). Preoperative and postoperative computed tomography angiography was used to measure the diameters of the thoracodorsal artery (TDA) and internal mammary artery (IMA) on the affected and unaffected sides. Differences in vessel diameters between sides and among groups were analyzed. Results: In groups B and D, the diameters of the affected TDA and IMA were significantly decreased compared with the changes observed on the unaffected side (p < 0.001). In contrast, there were no significant differences in vessel diameters between the affected and unaffected sides in groups A and C (group A: p = 0.644; group C: p = 0.367). Conclusions: Recipient vessel diameters for microsurgical breast reconstruction significantly decreased in patients who received postoperative RTx, with or without CTx. Plastic surgeons planning delayed breast reconstruction should be aware of these adjuvant therapy-related changes in recipient vessels and consider preoperative imaging assessment to accurately counsel patients regarding surgical risks and to support informed decision-making. Full article

Background: The aim of this retrospective study was to evaluate correlations between Optical Coherence Tomography (OCT) and Adaptive Optics (AO) of selected retinal parameters in individuals with increased BMI (≥25.0), including a subgroup analysis for hypertension (HTN). Methods: Sixty-three patients (120 eyes) were assessed using AngioVue OCT and rtx1TM AO devices. Retinal thickness (RT), optic nerve head (ONH), ganglion cell complex (GCC), retinal nerve fiber layer (RNFL), and photoreceptor (cone) parameters—density, spacing, regularity, dispersion—were analyzed. Results: A negative correlation between BMI and RT in the parafoveal superior and inferior quadrants was observed. Higher BMI was associated with thinner GCC in the superior and nasal parafoveal regions. Additionally, age negatively correlated with cone density and regularity, and positively with cone spacing and dispersion. Numerous correlations were noted between GCC values in OCT and cone parameters in AO, consistent across both HTN and non-HTN subgroups. Conclusions: The findings suggested that AO may detect retinal changes earlier than OCT. Multimodal imaging provides valuable insights into early structural changes associated with elevated BMI. Long-term monitoring is recommended to evaluate the progression and clinical impact of these findings. Full article

Magnetic resonance imaging (MRI) super-resolution (SR) enables high-resolution reconstruction from low-resolution acquisitions, reducing scan time and easing hardware demands. However, most deep learning-based SR models are large and computationally heavy, limiting deployment in clinical workstations, real-time pipelines, and resource-restricted platforms such as low-field and portable MRI. We introduce CHARMS, a lightweight convolutional–Transformer hybrid with attention regularization optimized for MRI SR. CHARMS employs a Reverse Residual Attention Fusion backbone for hierarchical local feature extraction, Pixel–Channel and Enhanced Spatial Attention for fine-grained feature calibration, and a Multi-Depthwise Dilated Transformer Attention block for efficient long-range dependency modeling. Novel attention regularization suppresses redundant activations, stabilizes training, and enhances generalization across contrasts and field strengths. Across IXI, Human Connectome Project Young Adult, and paired 3T/7T datasets, CHARMS (~1.9M parameters; ~30 GFLOPs for 256 × 256) surpasses leading lightweight and hybrid baselines (EDSR, PAN, W2AMSN-S, and FMEN) by 0.1–0.6 dB PSNR and up to 1% SSIM at ×2/×4 upscaling, while reducing inference time ~40%. Cross-field fine-tuning yields 7T-like reconstructions from 3T inputs with ~6 dB PSNR and 0.12 SSIM gains over native 3T. With near-real-time performance (~11 ms/slice, ~1.6–1.9 s per 3D volume on RTX 4090), CHARMS offers a compelling fidelity–efficiency balance for clinical workflows, accelerated protocols, and portable MRI. Full article

Background: Aggregatibacter actinomycetemcomitans is a Gram-negative, facultative anaerobic, immobile oral bacterium responsible for the secretion of virulence factors, namely leukotoxin (LtxA), a large exotoxin of the RTX family that enables the bacterium to evade the immune system by destroying leukocytes, resulting in aggressive periodontitis (AP) leading to tooth loss. Methods: This study aimed to screen 106 molecules derived from Moroccan propolis in order to identify potential inhibitors of the active sites of LtxA based on molecular docking, ADMET property evaluation, and molecular dynamics (MD) simulation. Results: Epigallocatechin gallate (EGCg), used as a reference compound, showed binding energies of −6.9 kcal/mol, −6.1 kcal/mol, −6.5 kcal/mol, and −5.9 kcal/mol with the four active sites P1, P2, P3, and P4, respectively. By establishing conventional hydrogen bonds, pi-alkyl bonds, and non-covalent pi–pi bonds. Chrysin and luteolin showed favorable binding affinities with the four active sites, named as follows: P1–P4 (P1–chrysin = −7.5 kcal/mol; P2–chrysin = −7.9 kcal/mol; P3–chrysin = −8.1 kcal/mol; P4–chrysin = −6.9 kcal/mol; P1–luteolin = −7.3 kcal/mol; P2–luteolin = −7.6 kcal/mol; P3–luteolin = −8.1 kcal/mol; P4–luteolin = −7.3 kcal/mol). The binding affinity of these two propolis derivatives was stabilized by pi−sigma bonds, pi−alkyl bonds, conventional hydrogen bonds, pi-cation interactions, non-covalent pi–pi bonds, and carbon–hydrogen bonds. According to free energy calculations performed with Prime MM-GBSA, the complexes formed by chrysin demonstrated the most stable interactions due to Van der Waals and lipophilic forces. Luteolin formed significant interactions, but slightly weaker than those of chrysin. These results reveal the inhibitory potential of chrysin and luteolin with protein active sites. MD simulations corroborated the excellent stability of complexes formed by chrysin, as indicated by low RMSD values, suggesting favorable dynamic behavior. Conclusions: These results highlight the potential of chrysin as a versatile inhibitor capable of interacting with the four active sites. These findings are a strong foundation for further experimental confirmations. Full article

Objective: Evaluate the efficacy and safety following intravitreal anti-vascular endothelial growth factor (anti-VEGF) therapy in patients with diabetic macular oedema (DME). Methods: To evaluate retinal microvascular remodelling and photoreceptor metrics using adaptive optics (AO) alongside systemic vascular status assessed by brachial/aortic hemodynamic and carotid ultrasound. We conducted a single-centre longitudinal study including twenty-one patients with DME. The following four diagnostic visits were performed: baseline (V1, no anti-VEGF treatment), 2–3 months (V2), 6–8 months (V3), and 12–14 months (V4). Adaptive optics (rtx1) measured foveal cone number (N) and regularity (Reg) within a standardised 80 × 80 µm window, and superior temporal retinal arteriole morphology after the first bifurcation (vessel diameter [VD], lumen diameter [LD], wall thickness [WT], wall-to-lumen ratio [WLR], and wall cross-sectional area [WCSA]). SphygmoCor provided peripheral (brachial) and central (aortic) pressures, augmentation pressure (AP), augmentation index (AIx), and carotid–femoral pulse wave velocity (PWV and PWVHR heart rate adjusted). Carotid ultrasound assessed intima–media thickness (IMT), carotid lumen diameter (CLD), and IMT/CLD ratio (IMTLR) 2 mm proximal to the bifurcation in diastole. Visual acuity (Visus), intraocular pressure (IOP), and central retinal thickness (CRT) were obtained at each visit. Results: In the treated eye (TE), WLR showed a significant overall change (Friedman p = 0.007), with a modest V4 vs. V1 increase (Wilcoxon p = 0.045); LD also varied across visits (Friedman p = 0.034). Cone metrics improved as follows: Reg increased over time (Friedman p = 0.019), with a significant rise at V4 vs. V1 (p = 0.018), and cone number increased at V3 vs. V1 (p = 0.012). Functional/structural outcomes improved as follows: visual acuity increased at V3 (p = 0.009) and V4 (p = 0.028), while CRT decreased at V3 (p = 0.002) and V4 (p = 0.030); IOP remained stable compared to V1. Systemic hemodynamics was largely unchanged; small fluctuations in DBP and cDBP across V1–V4 were observed (Friedman p = 0.034 and p = 0.022, respectively), whereas AIx, AP, PWV, and PWVHR showed no significant trends. Carotid IMT, CLD, and IMTLR did not change significantly across visits, supporting systemic vascular safety. Conclusions: Intravitreal anti-VEGF therapy in DME was associated with improvements in photoreceptor organisation and macular structure/function, with AO-derived arteriolar remodelling detectable over time, and no adverse changes in large-artery structure. These findings support ocular efficacy and systemic vascular safety; confirmation in larger cohorts is warranted. Full article

The explosive growth of artificial intelligence (AI) services has led to massive scaling of GPU computing clusters, causing sharp rises in power consumption and carbon emissions. Although hardware-level accelerator enhancements and deep neural network (DNN) model compression techniques can improve power efficiency, they often encounter deployment barriers and risks of accuracy loss in practice. To address these issues without altering hardware or model architectures, we propose a novel Carbon-Aware Resource Management (CA-RM) framework for GPU clusters. In order to minimize the carbon emission, the CA-RM framework dynamically adjusts energy usage by combining real-time GPU core frequency scaling with intelligent workload placement, aligning computation with the temporal availability of renewable generation. We introduce a new metric, performance-per-carbon (PPC), and develop three optimization formulations: carbon-constrained, performance-constrained, and PPC-driven objectives that simultaneously respect DNN model training deadlines, inference latency requirements, and carbon emission budgets. Through extensive simulations using real-world renewable energy traces and profiling data collected from NVIDIA RTX4090 GPU running representative DNN workloads, we show that the CA-RM framework substantially reduces carbon emission while satisfying service-level agreement (SLA) targets across a wide range of workload characteristics. Through experimental evaluation, we verify that the proposed CA-RM framework achieves approximately 35% carbon reduction on average, compared to competing approaches, while still ensuring acceptable processing performance across diverse workload behaviors. Full article

Although fine-tuning pretrained language models has brought remarkable progress to zero-shot named entity recognition (NER), current generative approaches still suffer from inherent limitations. Their autoregressive decoding mechanism requires token-by-token generation, resulting in low inference efficiency, while the massive parameter scale leads to high computational and deployment costs. In contrast, span-based methods avoid autoregressive decoding but often face large candidate spaces and severe noise redundancy, which hinder efficient entity localization in long-text scenarios. To overcome these challenges, we propose an efficient Embedding-based NER framework that achieves an optimal balance between performance and efficiency. Specifically, the framework first introduces a lightweight dynamic feature matching module for coarse-grained entity localization, enabling rapid filtering of potential entity regions. Then, a hierarchical progressive entity filtering mechanism is applied for fine-grained recognition and noise suppression. Experimental results demonstrate that the proposed model, which is trained on a single RTX 5090 GPU for only 24 h, attains approximately 90% of the performance of the SOTA GNER-T5 11B model while using only one-seventh of its parameters. Moreover, by eliminating the redundancy of autoregressive decoding, the proposed framework achieves a 17× faster inference speed compared to GNER-T5 11B and significantly surpasses traditional span-based approaches in efficiency. Full article

Repeat-in-toxin (RTX) toxins are calcium-dependent exoproteins secreted by diverse Gram-negative bacteria and play central roles in cytotoxicity, immune modulation, and tissue colonization. While their structure and secretion mechanisms are well-characterized, the regulation of RTX toxin expression remains complex and species-specific. This review provides a comprehensive overview of the regulatory networks governing RTX gene expression, highlighting both conserved mechanisms and niche-specific adaptations. RTX genes are controlled by multilayered regulatory systems that integrate global transcriptional control, metabolic regulation, and environmental sensing. Expression is further shaped by host-derived signals, physical contact with host cells, and population-dependent cues. Quorum sensing, post-transcriptional regulation by small RNAs, and post-translational activation mechanisms contribute additional layers of control to ensure precise regulation of toxin production. We also explore how RTX regulation varies across anatomical niches, including the gut, lung, bloodstream, and biofilms, and how it is co-regulated with broader bacterial virulence. Finally, we discuss emerging insights from omics-based approaches and the potential of anti-virulence strategies targeting RTX regulatory pathways. Together, these topics underscore RTX regulation as a model for adaptive virulence control in bacterial pathogens. Full article

Connective tissue disease-associated interstitial lung disease (CTD-ILD) comprises a heterogeneous group of immune-mediated pulmonary disorders with significant morbidity and mortality. The pathogenesis involves complex interactions of autoimmunity, chronic inflammation, and fibrosis. B cells play a central role in these processes through antigen presentation, autoantibody production, cytokine secretion, and the formation of ectopic lymphoid tissue within the lung parenchyma. Rituximab (RTX)—a chimeric anti-CD20 monoclonal antibody—depletes B cells and has emerged as a promising therapeutic agent for CTD-ILD. This review comprehensively presents the immunopathogenic mechanisms underlying CTD-ILD, elaborating on the multifaceted mode of action of RTX and summarizing the evolving clinical evidence. Full article

In dynamic scattering media (such as rain, fog, biological tissues, etc.) environments, scattered light causes severe degradation of target images, directly leading to a sudden drop in the detection confidence of target detection models and a significant increase in the rate of missed detections. This is a key challenge in the intersection of optical imaging and computer vision. Aiming to address the problems of poor generalization and slow reasoning speed of existing schemes, we construct an end-to-end framework of multi-stage preprocessing, customized network reconstruction, and object detection based on the existing network framework. First, we optimize the original degraded image through preprocessing to suppress scattered noise from the source and retain the key features for detection. Relying on a lightweight and customized network (with only 8.20 M of parameters), high-fidelity reconstruction is achieved to further reduce scattering interference and ultimately complete target detection. The reasoning speed of this framework is significantly better than that of the existing network. On RTX4060, the network’s reasoning ability reaches 147.93 frames per second. After reconstruction, the average confidence level of dynamic object detection is 0.95 with a maximum of 0.99, effectively solving the problem of detection failure in dynamic scattering media. It can provide technical support for scenarios such as unmanned aerial vehicle (UAV) monitoring in foggy weather, biomedical target recognition, and low-altitude security. Full article

This work establishes a large language model (LLM) specialized in the domain of thermoelectric generators (TEGs), for deployment on local hardware. Starting with the generalist JanV1-4B model and Qwen3-4B-Thinking-2507 models, an efficient fine-tuning (FT) methodology using quantized low-rank adaptation (QLoRA) was employed, modifying only 3.18% of the total parameters of thee base models. The key to the process is the use of a custom-designed dataset, which merges deep theoretical knowledge with rigorous instruction tuning to refine behavior and mitigate catastrophic forgetting. The dataset employed for FT contains 202 curated questions and answers (QAs), strategically balanced between domain-specific knowledge (48.5%) and instruction-tuning for response behavior (51.5%). Performance of the models was evaluated using two complementary benchmarks: a 16-question multilevel cognitive benchmark (94% accuracy) and a specialized 42-question TEG benchmark (81% accuracy), scoring responses as excellent, correct with difficulties, or incorrect, based on technical accuracy and reasoning quality. The model’s utility is demonstrated through experimental TEG design guidance, providing expert-level reasoning on thermal management strategies. This study validates the specialization of LLMs using QLoRA as an effective and accessible strategy for developing highly competent engineering support tools, eliminating dependence on large-scale computing infrastructures, achieving specialization on a consumer-grade NVIDIA RTX 2070 SUPER GPU (8 GB VRAM) in 263 s. Full article

The use of convolutional neural networks in nursery production remains limited, emphasizing the need for advanced vision-based approaches to support automation. This study evaluated the feasibility of detecting chip-budding graft sites in apple nurseries using YOLO object detection and segmentation models. A dataset of 3630 RGB images of budding sites was collected under variable field conditions. The models achieved high detection precision and consistent segmentation performance, confirming strong convergence and structural maturity across YOLO generations. The YOLO12s model demonstrated the most balanced performance, combining high precision with superior localization accuracy, particularly under higher Intersection-over-Union threshold conditions. In the segmentation experiments, both architectures achieved nearly equivalent performance, with only minor variations observed across evaluation metrics. The YOLO11s-seg model showed slightly higher Precision and overall stability, whereas YOLOv8s-seg retained a small advantage in Recall. Inference efficiency was assessed on both high-performance (RTX 5080) and embedded (Jetson Orin NX) platforms. YOLOv8s achieved the highest inference efficiency with minimal Latency, while TensorRT optimization further improved throughput and reduced Latency across all YOLO models. These results demonstrate that framework-level optimization can provide substantial practical benefits. The findings confirm the suitability of YOLO-based methods for precise detection of grafting sites in apple nurseries and establish a foundation for developing autonomous systems supporting nursery and orchard automation. Full article