Search Results (1,225)

Background/Objectives: Large language models (LLMs) are increasingly used by the public to seek health information, yet their accuracy in addressing common vaccine myths remains unclear. Sycophantic LLM behavior, where models align with rather than correct user-stated beliefs, poses specific risks in health contexts. Methods: We conducted an exploratory multi-vendor evaluation of three LLMs (GPT-5, Gemini 2.5 Flash, Claude Sonnet 4) using officially curated vaccination myths from Germany’s public health institution and two realistic user framings (curious skeptic, convinced believer). All model responses were independently evaluated by two blinded medical experts for misconception addressal (binary criterion applied to the response text), scientific accuracy, and communication clarity (5-point Likert scales). Additionally, blinded marketing experts ranked models for lay communication clarity. Flesch Reading Ease scores were computed for all outputs. Results: Across all myths, framings, and models (66 response items), both medical raters judged that all responses refuted the targeted misconception; no response affirmed or ignored a myth, including under the adversarial convinced believer framing. Scientific accuracy and clarity ratings were high and tightly clustered (median 4.0–4.5), with no combined score below 3 and substantial inter-rater agreement. Marketing experts independently ranked Gemini 2.5 Flash and GPT-5 highest for lay clarity. Readability analysis revealed generally low accessibility, particularly for the convinced believer framing and for Claude Sonnet 4 outputs. Conclusions: Our findings suggest that general-purpose LLMs can produce scientifically accurate, on-topic rebuttals to widely documented vaccine myths under realistic default conditions, although linguistic complexity and framing-sensitive style may limit accessibility. Whether such outputs change beliefs or behavior in hesitant individuals was not tested. With readability optimization, these outputs could serve as building blocks for myth-debunking tools, given prospective evaluation with behavioral endpoints. Full article

Aiming at core bottlenecks in weak and small target detection in radar range-Doppler (RD) spectra under low signal-to-noise ratio (SNR)—including severe performance degradation of traditional constant false alarm rate (CFAR) detectors and the inherent trade-off difficulty faced by existing deep learning methods in balancing detection accuracy, localization precision, and real-time performance—this paper proposes a progressive sub-pixel-level intelligent detection algorithm named F²DN-CCWL. The algorithm constructs a three-stage detection pipeline: global candidate screening, local fine discrimination, and weighted localization, and implements a full-stack customized design covering network architecture, soft-label training strategy, and post-processing modules. Simulation and field-measured results demonstrate that at −20 dB SNR, the proposed algorithm achieves a detection probability of 95.3%, a false alarm rate of 3.1%, an average localization error of 0.76 pixels, and a single-frame inference latency of 47.21 ms. This method offers a high-performance engineering solution for radar-based detection of low observable targets. Full article

To address the issues of low detection accuracy and high miss rates in long-range small traffic sign detection, which are caused by insufficient feature information and susceptibility to background interference, this paper proposes an improved real-time traffic sign detection algorithm based on YOLO11n. First, a cross-guided feature extraction module, C3k2_CGPEMA, is designed within the neck network. By embedding the Efficient Multi-Scale Attention (EMA) mechanism into the feature extraction branch of Partial Convolution (PConv), this module utilizes the spatial attention mask generated by the convolutional branch to provide cross-branch guidance and filter out complex background noise from the identity branch. This achieves precise fine-grained feature focusing while preserving high-frequency spatial details. Furthermore, a joint bounding box regression loss function combining Complete Intersection over Union (CIoU) and Gaussian Combined Distance (GCD) is adopted. This preserves the stable convergence properties of CIoU while leveraging the scale invariance of GCD to enhance the regression accuracy for small targets. Finally, the detection layers are reconstructed by removing the P5 layer and introducing a high-resolution P2 layer (160 × 160), significantly strengthening the localization capability for distant, tiny targets. Experimental results demonstrate that the proposed algorithm achieves improvements of 5.4, 7.4, and 6.6 points in precision, recall, and mAP@0.5, respectively, on the TT100K dataset compared to the baseline YOLO11n. While boosting detection accuracy, the model maintains an inference speed of 114.5 frames per second (FPS), fully satisfying the requirements for real-time detection in in-vehicle environments. Generalization experiments conducted on the CCTSDB dataset further validate the robustness of the proposed algorithm in complex environments. Full article

Real-time strawberry volume estimation is a pivotal technology for automated harvesting and precision grading. However, conventional contact methods are prone to damaging fruits, while existing vision-based approaches struggle to balance high accuracy with low computational overhead. To address these challenges, this study proposes a two-stage real-time volume estimation framework coupling a red-green-blue-depth (RGB-D) sensor with an “Instance segmentation–Principal-axis slicing” framework. First, to precisely extract target contours in complex backgrounds, we designed Deformable Feature Aware-YOLO (DFA-YOLO) based on the YOLO11-seg architecture. This model enhances the geometric perception of irregular fruit edges and effectively overcomes the challenges of background noise and multi-scale variations, providing high-precision masks for subsequent spatial mapping. Subsequently, a principal-axis-slicing algorithm extracts the mask’s centroid and principal axis, perpendicularly slicing the mask into infinitesimal micro-slices. By computing and accumulating the pixel-space volume of these slices, the system converts them into precise 3D physical volumes based on RGB-D depth mapping. The entire system was deployed on an NVIDIA Jetson Orin edge computing platform and validated in a greenhouse. Experimental results demonstrate that the estimated volume highly agrees with the true volume, achieving a coefficient of determination (R²) of 0.945 and a mean absolute percentage error (MAPE) of 9.0%. Under typical operating conditions (1–5 targets per field of view), the system maintains an overall frame rate of 8–15 FPS, requiring only 55 ms for single-fruit estimation. This method exhibits favorable stability and lightweight efficiency under the tested greenhouse conditions, offering a reliable solution for real-time non-destructive crop phenotypic monitoring in computationally constrained agricultural environments. Full article

Snapshot compressive imaging (SCI), including coded aperture compressive temporal imaging (CACTI), reconstructs high-speed video frames from compressed low-frame-rate measurements. Most deep SCI reconstruction networks are designed around a measurement-wise formulation: each compressed exposure is reconstructed independently, and the resulting frame segments are concatenated to form a continuous video. This protocol is effective for within-measurement reconstruction, but it leaves cross-measurement temporal continuity largely unmodeled. Boundary artifacts such as flickering, texture drift, or motion jumps can therefore appear between adjacent reconstructed segments, even when frame-wise reconstruction metrics remain competitive. This work identifies and empirically analyzes the underexplored problem of cross-measurement temporal continuity in continuous SCI, and it provides TCA-EfficientSCI as a lightweight, causal, and reproducible baseline. The Temporal Context Adapter uses the last m reconstructed frames from the previous measurement as causal temporal context and injects this history through a gated residual feature pathway. A boundary consistency loss regularizes the predicted temporal variation across measurement boundaries without forcing adjacent frames to be identical. In a controlled three-seed comparison, Full TCA with boundary loss reduces mean Boundary Difference Error (BDE) by 2.23% relative to the matched-epoch EfficientSCI control while maintaining similar PSNR and SSIM. Correct-history inference gives BDE 0.01615, while zero and shuffled history give 0.01725 and 0.01810, respectively. The adapter adds 1,019,905 parameters, or 11.56% relative to the EfficientSCI baseline parameters, and it changes $256\times 256$ mean latency from 54.35 ms to 68.58 ms per measurement in the profiling protocol. Rather than claiming broad reconstruction-quality improvement, this study highlights cross-measurement continuity as an important evaluation and design dimension for continuous SCI deployment. Full article

We introduce the Hamiltonian pseudorandom function (HPRF), a new symmetric cryptographic primitive in which the function family $\left\{F_k\right\}$ is defined by $F_k\left(q\right)=\nabla S_k\left(q\right)$, the gradient of the generating function of a secret Lagrangian submanifold $L_k$ on the symplectic torus ${\mathbb{T}}^{2n}$. The key k specifies a composition of kicked-rotor maps in the strongly chaotic regime, whose classical Lyapunov exponents grow as $log(K/2)$ per kick. The HPRF is best understood as a seeded one-way function with high min-entropy output: $F_k$ is smooth ($C^{\infty }$), so its raw output is not directly usable as a uniform keystream, but it is computationally hard to invert. We construct three symmetric encryption modes—Mode A (key-dependent coordinate frame), Mode C (Lagrangian keystream), and Mode AC (hybrid)—in which the HPRF supplies the hardness and a key derivation function (HKDF) supplies bit-level uniformity. Standard symmetric composition then yields IND-CPA and IND-CCA2 security. Classical security reduces to the Lagrangian identification problem (LIP), shown as equivalent to the Hamiltonian inversion problem of recovering the kick parameters, which we state as an explicit hardness assumption supported by a precision/sample-complexity obstruction from the positive Lyapunov exponents, by the empirical failure of concrete attacks, and (more heuristically) by topological suggestiveness from the Arnold conjecture and Floer theory. We validate a gradient-fitting attack and an algebraic-structure attack and show that both fail. For quantum security, we propose what we believe is the right framing: that the composed Floquet operator ${\widehat{U}}_K^r$ is a candidate pseudorandom unitary (PRU) in the sense of Ji–Liu–Song. We provide three independent pillars of evidence—Wigner–Dyson spectral statistics, Lyapunov-rate scrambling, and conjectural approximate-design behaviour—and reduce the HPRF quantum security to the PRU conjecture for ${\widehat{U}}_K^r$. We then retire the dynamical-localisation argument of previous drafts as inapplicable at cryptographic parameters; the chaotic-pseudorandomness regime that the operator actually inhabits is, we argue, a stronger foundation than the one that localisation would have provided. A deterministic fixed-point arithmetic core ensures cross-platform bit-exact consistency. A reference implementation validates correctness across all modes, and an NIST SP 800-90B analysis of the output min-entropy fixes the parameter sets. As a foundational proposal, the HPRF is intended for settings that seek a symmetric hardness assumption structurally independent of the algebraic problems underlying current cryptography, for example, as a hedge primitive in defence-in-depth designs, or as a basis for further study of geometry- and chaos-based cryptography, rather than as a drop-in replacement for AES or lattice-based schemes at this stage. Full article

Real-time semantic segmentation models offer an excellent balance between accuracy and inference speed. However, deploying these models in dynamic real-world environments often requires the ability to learn novel classes incrementally without retraining on the entire dataset. This capability is known as continual learning. In this regard, standard fine-tuning methods often suffer from catastrophic forgetting, where the model learns new information but loses accuracy on previously learned classes. The severity of this effect depends on the incremental setup, the available data, and the fine-tuning strategy. Contributing to this crucial domain, this paper proposes a novel continual learning framework tailored for PIDNet, which is a widely cited state-of-the-art real-time semantic segmentation model. Our method, PILOT (Parallel Incremental Learning Over Time), introduces a real-time and lightweight strategy by implementing a parallel Derivative branch (D-branch) designed to capture the high-frequency boundary information of novel classes while freezing the trained parameters of the original segmentation network. This novel setup allows the model to adapt to new semantic categories while preserving the knowledge of previously learned classes. By using only data associated with the new class, our model significantly reduces training overhead. Experimental results demonstrate that our approach successfully segments new classes while maintaining a high mean Intersection over Union (mIoU) on the original base classes, thereby outperforming prior continual learning approaches in this real-time segmentation setting. Overall, PILOT is shown to effectively mitigate catastrophic forgetting with minimal impact on inference latency, adding fewer than 5% additional parameters and reducing the frame rate by only about 9%, thus maintaining real-time performance. Full article

The worldwide spread of semiconductor devices has driven a surge in electronic waste (e-waste), which reached 62 million metric tons in 2022 and is projected to exceed 80 million metric tons by 2030. E-waste contains hazardous substances such as cadmium and mercury, yet also represents a $57 billion annual opportunity through the recovery of valuable and critical raw materials (CRMs). However, formal recycling rates remain stagnant at 22.3%, largely due to limitations of current automated sorting methods. These systems primarily rely on visible-light (RGB) imaging, which lacks the spectral resolution needed to distinguish chemically similar polymers, complex metal alloys, and composite substrates on printed circuit boards (PCBs). This paper presents a multidisciplinary synthesis of AI-driven detection and classification for e-waste, bridging materials science and computer vision through three interconnected themes. 1. Material and Economic Context: The toxicological risks and economic drivers of semiconductor recycling are characterized, framing fine-grained material identification as essential for a circular economy. 2. Multispectral Sensing & Fusion: Sensing modalities such as near-infrared (NIR), hyperspectral imaging (HSI), and X-ray fluorescence (XRF) are assessed, and sensor fusion strategies, including early, late, and intermediate fusion, are reviewed for high-throughput industrial settings. 3. Deep Learning Benchmarking: 11 publicly available PCB datasets are analyzed, and the YOLO series (YOLOv3–YOLOv12) is compared with leading non-YOLO detectors, including Faster R-CNN, RT-DETR-L, and RetinaNet. The results show that while YOLOv9s achieves a peak mAP@0.5 of 56.5% and YOLOv11s offers an optimal industrial profile (37.2% mAP@0.5:0.95 at 115 ms edge inference), all RGB-based models fail to detect visually ambiguous surface-mount devices (SMDs), with mAP values below 12%. This confirms a performance ceiling for purely visual systems. The review concludes that transitioning from RGB-centric to multispectral fusion architectures is the primary research frontier and proposes a roadmap for standardized multimodal datasets and edge-deployable fusion models to enable next-generation, high-recovery automated recycling. Full article

Scene matching-based localization systems (SMLSs) offer an effective solution to the failure of Global Navigation Satellite System (GNSS) positioning in complex environments. This paper designs and implements a vision-based autonomous localization system for unmanned aerial vehicles (UAVs), compatible with both optical and infrared sensors, delivering high frame rates and high-precision positioning performance. First, to address the issue of uneven texture distribution in natural terrain features, an adaptive expansion sliding window model is constructed to accurately extract texture-rich regions, which effectively improves matching precision. Second, considering the differences in edge characteristics between optical and infrared images, the Sobel operator and Scharr operator are introduced, respectively, to construct gradient features, achieving high-precision, high-frame-rate heterogeneous image matching. Furthermore, to significantly improve the system frame rate, this paper designs an embedded parallel acceleration strategy based on a multi-core CPU architecture. The strategy achieves task-level concurrency between the front-end stages (pre-correction and feature refinement) and matching, and implements parallel optimization for feature construction and correlation computation within the matching module. On the algorithmic level, the correlation computation is further accelerated by replacing spatial-domain convolution with frequency-domain multiplication. Finally, the algorithm is deployed on an RK3588 embedded platform. The effectiveness of the proposed system is validated using offline flight data from a medium-altitude fixed-wing UAV and real-time flight experiments with a low-altitude rotary-wing UAV. In the medium-altitude UAV flight data validation, optical visual localization achieves an average position error of 20.94 m with a processing time of 0.123 s/frame, while infrared visual localization yields a position error of 11.77 m at 0.128 s/frame. In the low-altitude UAV flight experiment, optical visual localization achieves an average position error of 9.68 m at 0.15 s/frame, and infrared visual localization achieves an average position error of 11.22 m at 0.15 s/frame. Full article

Tool-calling LLM agents are vulnerable to indirect prompt injection: externally retrieved data can redirect tool calls without system-prompt access, and prompt-level defences leave three harm classes undefended (path traversal, user-guided exfiltration, high-frequency tool abuse). We present a Policy Enforcement Point (PEP) that intercepts at the tool-call boundary with declarative rules over a cross-step information-flow label system (source integrity, data sensitivity) and a synchronous SHA-256 hash-chained audit log. On a controlled dataset across four attack classes, the full system cuts the attack success rate (ASR) from 40.0% to 5.0% (deepseek-v4-pro, five repeats) versus 35.0% for the strongest prompt-only baseline; disabling cross-step label propagation raises the call-level false-negative rate by 26.4 points. The 30.0% task-level false-positive rate is dominated by by-design least-privilege capability-token denials, not rule false positives—an expanded 30-task benign set yields 0/30 rule false positives under scripted isolation. A conservative-DS mitigation (intent-taint) closes the constructed denied-read reconstruction blind-spot variant (ASR 100% to 0%) at no cost on standard workflows. The audit log detects all three tested tamper classes; the in-process enforcement overhead is sub-millisecond per call. Across four further backends, ASR drops under the full system, though LLaMA-3.3-70B retains 16.7% (a rule-coverage gap). A preliminary run over a real MCP stdio transport (an official filesystem server) shows the mechanism operates at a real boundary with a sub-millisecond execution-path increment. We frame these as mechanism-coverage evidence on a controlled benchmark, not a deployability claim for production MCP workloads. Code, data, and metrics are openly available in the replication repository. Full article

To address the challenge of high false alarm rates caused by dynamic viewpoint noise in mobile chemical inspections, this study established a highly robust adaptive dynamic risk inference model. This research proposes an inference framework integrating spatio-temporal semantic constraints. Spatially, this study constructed a heterogeneous dynamic scene graph and introduced a kinematic-aware anisotropic dynamic field. This field transforms geometric hard boundaries into continuous risk gradients that deform dynamically with target intentions to suppress observation ambiguity. Temporally, the work designed an uncertainty-aware adaptive hysteresis filter, whose state machine thresholds adjust dynamically according to real-time sensor noise levels. Comparative tests on a real-world chemical dataset show that the model achieves a peak F1-Score of 93.1%, reduces the false alarm rate to 1.3 times/h, and requires a single-frame processing time of only 24.8 ms. The method theoretically achieves spatio-temporal dynamic noise reduction, significantly mitigates topological mutations and alarm chattering under complex visual noise conditions, meets edge computing deployment requirements, and provides a high-confidence sensing decision hub for industrial process safety monitoring. Full article

High-speed imaging offers a non-intrusive approach for monitoring falling-film flows over horizontal tube bundles, but reflective images are difficult to quantify because grayscale variations are jointly affected by film geometry, interfacial curvature, surface slope, viewing angle, and local highlights. This study proposes an interpretable visual-proxy sensing framework for comparative state assessment of such flows. Isothermal water experiments were conducted on a five-row horizontal tube bundle over Re_Γ = 184 − 960. For each condition, grayscale frames were acquired at fps and analyzed within five fixed row-wise regions of interest. The image sequence was transformed by temporal-median background subtraction, local spatiotemporal mapping, moving-average detrending, and median-absolute-deviation normalization. The resulting normalized map M_n and dynamic renewal field G were used to extract four scalar descriptors: noise-corrected apparent renewal intensity I_R, high-frequency fraction R_HF, spectral peak frequency f_p, and burst-event rate F_B. Results show that M_n and G capture the transition from sparse column flow to more continuous sheet flow and reveal row-dependent activity organization. The descriptors provide complementary information on renewal intensity, frequency composition, dominant time scale, and intermittent events. Zero-response, noise-correction, and sensitivity tests confirm that the framework avoids structured pseudo-waves and maintains stable row-wise comparisons. The method provides a low-calibration visual sensing tool for relative falling-film state assessment. Full article

HCT116 is a colorectal cancer cell line frequently used in anti-tumor drug development experiments as well as in studies of the molecular machinery of eukaryotic cells. It is well characterized by the presence of several single-nucleotide and short mutations in multiple oncogenes and tumor suppressor genes, including KRAS, PIK3CA, MLH1, CTNNB1, CDKN2A, TGFBR2, and BRCA2. However, its landscape of large genomic rearrangements (LGRs) and copy number variants (CNVs) is still far from being fully understood. Therefore, the aim of this study was to identify LGRs and CNVs in several HCT116 cell line samples using Oxford Nanopore sequencing technology, including three samples from the SRA NCBI database, and to compare common and unique variants across all samples. Using the recently developed eLaRodON tool, we identified 22,666 common LGRs, among which more than 70% of tandem duplications and deletions larger than 80 kb were confirmed by CNV analysis. Among LGRs affecting protein-coding sequences, two in-frame rearrangements were identified: a deletion of exons 4–6 and a duplication of exon 10 in the CCSER1 gene, which encodes a cell division regulator protein. Given its high rearrangement rate in various tumors and the clinical significance of its overexpression, this finding may be potentially useful in future research on this cell line. Regarding differences between samples, we found that LGRs in the laboratory sample and in one of the three SRA NCBI samples occurred more frequently via ALR/Alpha repeats than via Alu repeats, in contrast to common LGRs and those unique to the other samples, a finding that may indicate the presence of unique mechanisms of genomic instability. Thus, this study reveals a broad spectrum of large genomic rearrangements and copy number variants that can be identified in the HCT116 cell line using Oxford Nanopore sequencing, including rearrangements specific to distinct cell line samples. Full article

Background and Objectives: As Romanian health systems deploy artificial intelligence (AI), uptake depends on navigating the EU AI Act, GDPR, the Medical Device Regulation (MDR), and national rules. We measured AI legislation literacy, governance readiness, and adoption intentions among Romanian healthcare professionals, identified implementation phenotypes, and tested whether confidence mediates the literacy–adoption link. Materials and Methods: In a multicenter cross-sectional survey (N = 109), participants completed a 20-item AI Legislation Literacy Index (0–20) plus scales rated form one to five measuring legislative confidence, adoption intention, readiness, trust, and perceived compliance burden. We used PCA and k-means clustering, multivariable logistic regression for high adoption intention (≥4), and covariate-adjusted mediation (5000 bootstrap resamples). Results: Mean age was 38.7 ± 9.8 years, and 60.6% of participants were female. Literacy was moderate (11.2 ± 4.1/20) and familiarity favored GDPR (69.7%) over the EU AI Act (25.7%). Literacy correlated with confidence (=0.52), whereas confidence correlated with adoption intention (=0.41); trust correlated positively (=0.44) and burden correlated negatively (=−0.29) with adoption. High adoption intention was noted in 50.5% of participants and was independently associated with higher literacy (aOR 1.85 per +1 SD; 95% CI 1.20–2.85), higher trust (aOR 1.72; 1.13–2.63), lower burden (aOR 0.64; 0.43–0.95), and prior AI training (aOR 2.10; 1.03–4.29). Three phenotypes emerged (Confident Adopters n = 44; Cautious Compliers n = 36; Skeptical Low Literacy n = 29), with adoption scores of 4.2 ± 0.5 vs. 3.1 ± 0.7 in the highest and lowest groups. Mediation showed a partial indirect effect via confidence (0.13; 95% CI 0.05–0.24). Conclusions: AI legislation literacy, confidence, trust, and perceived burden are key, modifiable determinants of AI adoption intentions; phenotype-guided strategies can target training, governance support, and post-deployment monitoring readiness. The revised framing explicitly situates these determinants within recent AI-specific regulatory and technical developments, including high-risk AI obligations, AI-enabled medical device change control, generative/large multimodal model risks, and lifecycle monitoring. Full article

Seashore paspalum (Paspalum vaginatum), a salinity-tolerant turfgrass, lacks well-characterized viral profiles. This study reports the discovery of a novel virus, tentatively named Paspalum latent virus (PaLV), representing a new taxon within the Alphaflexiviridae. Using high-throughput sequencing and RACE PCR, the 6995 nt genome was determined, revealing five open reading frames. Notably, PaLV lacks the AlkB domain and exhibits unique features, including overlapping start-stop codons (ORF4/ORF5) and a second in-frame AUG in the coat protein (CP) region. Phylogenetic analysis of the replicase placed PaLV in a distinct clade, separate from Potexvirus and Lolavirus. Despite low sequence identity, AlphaFold2 revealed conserved CP structural domains. Genetic analysis of 11 isolates showed low diversity and strong purifying selection. Pathogenicity assays through mechanical transmission demonstrated a broad but latent host range, including Zea mays and Sorghum spp. These findings suggest PaLV represents a novel species within a putatively new genus, Paspalovirus. Given its 90% incidence rate and latent profile, the RT-PCR assay developed here is vital for routine molecular diagnostics in turfgrass management and germplasm conservation. Full article