Search Results (1,033)

On edge devices, convolutional neural network (CNN) inference is bottlenecked mainly by memory bandwidth, owing to the frequent memory accesses to feature maps and parameters. To address this challenge, we propose a memory-efficient hardware accelerator for depthwise separable convolution that minimizes off-chip memory traffic and parameter storage. The proposed architecture employs three key techniques: (1) a 64-bit run-length coding (RLC) packet compression that exploits feature-map sparsity after ReLU, (2) a mixed-precision scheme that represents feature maps and weights at different precisions, and (3) separated depthwise and pointwise convolution units. In particular, feature maps are transferred in RLC-compressed form, which reduces the amount of data exchanged with the host. The compressed data are decoded row by row, so the on-chip buffers hold only the rows required for computation rather than a complete feature map. In software simulation on ImageNet, the mixed-precision scheme reduced the parameter storage by 49.22% at a cost of 6.24 percentage points (pp) in Top-1 accuracy, and the RLC reduced the data by up to 56.07% in the deeper layers. Implemented on a Xilinx ZCU-104 FPGA, the proposed accelerator performs the depthwise separable convolution with a small number of logic resources and on-chip memory, confirming its feasibility for resource-constrained edge devices. Full article

The STRYD^® power meter has been used by runners and running coaches in recent years. The data it provides may offer useful information for training monitoring and performance assessment. Accordingly, this systematic review identifies and analyzes the use of the main parameters provided by the STRYD^® power meter in the scientific literature. The PRISMA guidelines were used to identify relevant studies. The search was conducted in the electronic databases PubMed, Scopus and Google Scholar. In total, 142 studies were identified. After removing duplicates and assessing the eligibility of the abstracts, 29 studies were included in the systematic review. Eight of the main metrics provided by STRYD^® were analyzed. The total number of times each metric appeared varied widely, as did the number of metrics analyzed in each study. Power output was identified as the most frequently used metric in scientific studies and as the primary variable in the primary objectives sought. Ground contact time and cadence were also used by a high percentage of users, far exceeding the rest: vertical oscillation, flight time, stride length, form power, and stiffness. The study provides coaches and technical staff with guidelines on which variables to use when analyzing runners using STRYD^®. Full article

This study presents the development of an attribute control chart for monitoring the number of failures in a truncated life test under a single sampling plan, where the product lifetime follows either the Compound Weibull (CW) distribution or its special case, the Compound Exponential–Gamma (CEG) distribution. The control chart constants were determined for various parameter settings through an iterative design procedure. Specifically, the chart parameters were obtained for different quality levels and specified in-control Average Run Length (ARL₀) values. Furthermore, the performance of the proposed control charts was evaluated in terms of the out-of-control Average Run Length (ARL₁), Standard Deviation of Run Length (SDRL), Expected Average Run Length (EARL), and Extra Quadratic Loss (EQL) under various process shift levels. Numerical tables were provided for different combinations of the shape parameters, sample sizes, target ARL₀ values, and shift parameters. Finally, a simulation study was conducted to demonstrate the effectiveness of the proposed control charts in monitoring defective items. Overall, the results demonstrate that the proposed attribute control chart is applicable to processes in which product lifetimes follow the CW and CEG distributions and provides an effective tool for monitoring process shifts. The study demonstrates how the np chart’s performance is affected under different scenarios regarding CW and CEG distributions. It was observed that the chart exhibits high performance by enabling early detection during significant process shifts. In cases of minor process changes, the chart was found to be sensitive under the CW distribution assumption. Full article

This study investigated how detection distance affects spectral models for soluble solids content (SSC) and firmness evaluation in Korla fragrant pears and provides a reference for calibrating standardized indoor non-destructive detection equipment. Two hundred visually intact fruit samples at the early-ripening stage were collected from the Korla production area in Xinjiang. An FS-640 multispectral camera system equipped with a VS-SWR fixed-focus industrial lens (16 mm focal length, F1.8 maximum aperture, 1/2-inch sensor format) was used to acquire fruit reflectance spectra at seven vertical lens-to-fruit-surface distances of 90, 100, 110, 120, 130, 140, and 150 cm. A 625-pixel region of interest (ROI) was selected using ENVI at an undamaged equatorial or near-equatorial position of each fruit, and the regional mean spectrum was used as the spectral feature of one fruit sample. The sample-set partitioning based on joint X–Y distances (SPXY) algorithm was used to divide the calibration and prediction sets at a 3:1 ratio after outlier removal via a residual-threshold method. Four preprocessing methods, namely LOESS smoothing, standardization, vector normalization, and Savitzky–Golay (SG) smoothing, were compared. Competitive adaptive reweighted sampling (CARS) was performed with 50 Monte-Carlo sampling runs, a maximum of 30 principal components, and 10-fold cross-validation, yielding 99 characteristic wavelengths. Partial least squares regression (PLSR), support vector regression (SVR), random forest (RF), and artificial neural network (ANN) models were then established using identical input variables and sample partitions. Model performance was evaluated using the coefficient of determination for calibration ($R_c^2$), coefficient of determination for prediction ($R_P^2$), root-mean-square error of calibration (RMSEC), root-mean-square error of prediction (RMSEP), relative prediction deviation (RPD), and ratio of performance to interquartile distance (RPIQ). Under the static laboratory acquisition conditions in this work, the SSC model achieved the best prediction performance at 110 cm with SG smoothing ($R_P^2$) = 0.8949, RPD = 3.0633, RPIQ = 5.8661), whereas the firmness model obtained optimal prediction performance at 140 cm with standardization ($R_P^2$) = 0.7460, RPD = 1.9425, RPIQ = 3.2867). Changes in detection distance altered illumination uniformity, effective reflected signal, photon-scattering paths, and background-noise proportion. These effects may partially explain why the chemical-absorption-dominated SSC index and the tissue-scattering-dominated firmness index responded differently to detection distance. The results provide a reference for setting spectral detection parameters for Korla fragrant pears; however, samples were obtained from only a single producing region, harvest season, and maturity stage, and no independent external validation dataset was used. Therefore, the generalization ability of the developed models needs to be further verified using cross-season and cross-orchard sample sets. Full article

Sprint performance improves throughout adolescence as a result of both structural and neuromuscular development. In particular, increases in body height, lower limb length, and muscle volume have been consistently linked to improvements in sprint performance. However, allometric scaling of force, speed, and power remain lower in adolescents compared to adults even after structural differences are accounted for, implicating neural factors as independent contributors to performance development. The purpose of this study was to investigate differences in neuromuscular activation patterns of the lower limb muscles during maximal sprinting between youth male athletes across two age groups (U19: 17–19 years; U16: 13–16 years). Eighteen athletes performed a 50 m maximal sprint. Spatiotemporal variables (running speed, step frequency, step length) were measured over 30–50 m using a high-speed camera (240 Hz) and timing gates. Electromyographic (EMG) signals were recorded simultaneously from ten lower limb muscles using wireless EMG sensors (2000 Hz): rectus femoris (RF), biceps femoris (BF), semitendinosus (ST), gluteus maximus (Gmax), gluteus medius (Gmed), vastus lateralis (VL), vastus medialis (VM), tibialis anterior (TA), gastrocnemius (GAS), and soleus (SOL). Root mean square (RMS) amplitude was calculated across four gait phases (contact, early-swing, mid-swing, late-swing) and normalised to maximal voluntary Isometric contraction (%MVIC). The U19 group demonstrated significantly greater running speed (U19: 9.49 ± 0.39 vs. U16: 8.67 ± 0.25 m·s⁻¹, p < 0.001), step frequency (U19: 4.49 ± 0.12 vs. U16: 4.35 ± 0.16 Hz, p = 0.004), and step length (U19: 2.12 ± 0.12 vs. U16: 1.99 ± 0.06 m, p = 0.010) than U16. The overall pattern of lower limb muscle activation across the gait cycle was broadly similar between groups; however, a significant group × phase interaction was observed for RF (p = 0.003, F = 5.257, η² = 0.247), with post hoc analysis revealing greater RF activation during early swing in U19 (p = 0.033). These findings may indicate that sprint-specific training in youth athletes is associated with not only structural but also neuromuscular differences, specifically reflecting enhanced RF recruitment during the phase-critical moment of early swing—a window in which high-threshold motor unit activation is most mechanically decisive. EMG-based assessment of hip flexor activation during maximal sprinting may provide a complementary tool, pending further validation, for talent identification and training prescription in youth track and field. Full article

Masu salmon (Oncorhynchus masou) is an ecologically and economically important cold-water salmonid in East Asia that exhibits diverse life-history forms. To compare population-level genomic variation and patterns of homozygosity among fish from different sources, we analyzed whole-genome resequencing data from 465 individuals representing one field-collected Tumen River population (TM) and two landlocked cultured populations from Chicheng (CC) and Yanji (YJ). After quality control, 6,220,980 high-quality SNPs were retained. Population-specific filtering identified 5,589,828, 3,545,209, and 4,975,751 polymorphic SNPs in CC, TM, and YJ, respectively; although SNP numbers differed, approximately 91% of variants in each population were located in intronic or intergenic regions. Principal component analysis, ADMIXTURE, and distance-based neighbor-joining analysis clearly distinguished the three populations, with CC and YJ showing the closest genetic relationship. Pairwise $F_{\mathrm{ST}}$ was lowest between CC and YJ and highest between TM and YJ. CC exhibited the highest linkage disequilibrium, whereas TM showed the fastest LD decay and the lowest nucleotide diversity and heterozygosity. Runs of homozygosity (ROH) burden was highest in TM, intermediate in CC, and lowest in YJ. TM had the highest number of ROHs, cumulative ROH length, and $F_{\mathrm{ROH}}$, and ROHs longer than 5 Mb were detected only in this population. CC had an intermediate ROH burden dominated by short segments, whereas YJ had the lowest ROH-based genomic inbreeding. The high and heterogeneous ROH burden in TM indicates elevated genome-wide homozygosity among the sampled fish but does not, by itself, demonstrate recent inbreeding throughout the population. Candidate ROH islands and their annotated genes showed limited overlap among populations. Candidate genes in TM were primarily associated with ion regulation, neural processes, and energy metabolism, whereas those in CC and YJ shared broad functional categories involving development, muscle organization, nutrient transport, and neural regulation but differed in most specific genes. These regions and genes should be regarded as exploratory, hypothesis-generating candidates rather than evidence of selection or causality. Overall, this study reveals distinct population genomic characteristics and ROH patterns among masu salmon populations of different origins and provides a basis for future germplasm conservation and genetic management. Full article

Water scarcity, high non-productive soil evaporation, and poor vegetation establishment are major constraints on the sustainable ecological restoration of reconstructed open-pit mine dumps in arid and semi-arid regions. Conventional surface-applied water replenishment can result in rapid evaporative loss, thereby reducing the ecological benefits obtained from limited water resources. However, whether redistributing water into deeper reconstructed soil layers can simultaneously reduce non-productive evaporation, stabilize the root-zone hydrothermal environment, and improve vegetation growth remains insufficiently verified. In this study, a deep-profile soil water replenishment (DPSWR) device was tested in reconstructed mine-dump soil columns planted with locally adapted Stipa. Surface-applied water replenishment (CK) and DPSWR were compared using a single-run simulated rainfall comparison, soil water-retention and water-loss measurements, continuous temperature and moisture monitoring at 10 and 40 cm depths, and plant growth indicators. In the rainfall-simulation comparison, DPSWR showed lower cumulative water loss across the tested rainfall intensities and improved water-retention stability; the evaporation rate under CK was approximately 1.3 times that under DPSWR, whereas final soil water-holding capacity under DPSWR was approximately 2.4 times that under CK. Root fresh weight, plant fresh weight, and seedling number were significantly higher under DPSWR than under CK (p < 0.01), and maximum plant height and root length also increased significantly (p < 0.05). Under equal water-input conditions, DPSWR reduced non-productive water loss, prolonged soil water retention, and supported vegetation establishment. These findings suggest that DPSWR may provide a more water-efficient approach to the sustainable restoration of reconstructed mine dumps in water-limited regions. Full article

Developing autonomous navigation systems for small fishing vessels is required to improve path-tracking robustness and mitigate severe wave-induced roll motions. This study proposes a Deep Q-Network (DQN)-based operational path planning methodology that explicitly incorporates roll motion reduction into the reward function, combining a maneuvering model with hydrodynamic analyses. Simulation results under varying wave directions and heights demonstrate that the vessel actively adjusts its heading to minimize the roll response. Based on statistical evaluations across five independent runs, the proposed model effectively reduced the average and maximum roll responses by an average of 3% and 2%, respectively, under the evaluated wave headings at a wave height of 1.0 m, while maintaining operational path tracking, despite a slight increase in the total operational path length. Future research will focus on integrating complex environmental conditions with wind and current, and performing the model test for the validation of the established DQN model. Full article

In the single-depot multi-traveling salesman problem, traditional depot location methods often overlook task balance among traveling salesmen, leading to excessive load on certain units and compromising overall operational efficiency. To address this issue, this paper proposes an optimized depot location method based on clustering and multi-task balancing. The core contribution lies in the design of a multi-weight adaptive depot optimization method. This approach clusters city nodes into multiple groups through cluster analysis and dynamically synthesizes direction vectors using information such as the number of samples within each cluster and the convex perimeter. It iteratively optimizes depot locations, minimizing the total path length while enhancing workload balance across all traveling salesman routes. Additionally, a “divide-and-conquer” strategy decomposes the complex MTSP into multiple parallel TSP subproblems, which are then efficiently solved using Or-Tools. A comprehensive evaluation framework is introduced, incorporating Total-Sum distance, Min-Max distance, Workload Balance, Cluster separability, Robustness, and Running time. Experimental results on the TSPLIB standard dataset demonstrate that the proposed method exhibits significant advantages over various traditional clustering algorithms in both route optimization and route balancing, validating its effectiveness and practicality. The method’s robust performance provides a reliable solution for real-world applications such as logistics distribution, further highlighting its practical value. Experimental results show that the proposed method reduces the total travel distance and improves workload balance on multiple TSPLIB instances compared with conventional depot selection baselines. Full article

During the conceptual design phase, detailed geometric models are often unavailable, hindering energy-driven decisions for five-story office buildings. This paper proposes a geometry simplification strategy for office buildings in hot–humid regions using only length and width. Based on 130,976 geometric parameter combinations, stratified into an 80% selection set and a 20% validation set using four-dimensional parameter stratification, a standardized rectangular energy model is built, and EnergyPlus simulates orientations from 0° to 179° (1° step), totaling 23.6 million runs. Three simplification methods are compared: aspect ratio, floor area, and the proposed length–width combination. The length–width combination simplification strategy achieves an average relative energy deviation of 6.89% (6.87% on the held-out validation set), with a reduced maximum deviation on the held-out validation set (27.70% for the length–width method vs. 32.07% for the floor area method), thus meeting conceptual design accuracy requirements. Using this simplified model, the optimal orientation is identified as 0° (true south–north), accounting for 83.14% of cases. For the three representative geometric configurations examined, a near-optimal orientation band of ±15° around their respective optimal orientations is identified, within which the energy penalty remains below 1% of the minimum, providing designers with potentially flexible guidance that accommodates site-specific constraints without compromising energy performance. The orientation range 39–86° is not recommended when energy performance is prioritized, as it contains over 99% of worst-case orientations. The proposed strategy enables rapid energy estimation and orientation guidance from basic parameters, shifting energy-efficient design from late verification to early-stage driving, and providing quantifiable support for early-stage energy-efficient design as the foundation for subsequent nearly zero-energy building development in hot–humid regions. Full article

Background/Objectives: Large language models (LLMs) show promise for clinical decision support, yet their accuracy in interpreting specialized medical guidelines remains uncertain. Retrieval-augmented generation (RAG) may enhance performance by grounding responses in authoritative knowledge bases. This study aimed to compare the accuracy, comprehensiveness, and safety of RAG-enhanced versus standard LLMs for answering clinical questions derived from the German S3 guideline for oral cavity carcinoma. Methods: We conducted a prospective, single-blind benchmark study evaluating six LLMs: one RAG-enhanced model (Custom GPT with guideline access), one consensus-based model (ConsensusGPT), and four standard models (DeepSeek-V3.2, Mistral Small 3.2, Qwen3-Next-80B, GPT-OSS-120B). Fifty clinical questions covering 17 guideline domains were presented to each model three times, yielding 900 evaluations. Three expert reviewers assessed responses using 5-point Likert scales for accuracy, comprehensiveness, and clarity, under a single-blind procedure, the effectiveness of which was tested by a pre-specified manipulation check. We then ran a paired within-model experiment in which each base model was queried with and without guideline access through a transparent, openly released retrieval pipeline, and scored every response with a condition-blind automated judge alongside deterministic retrieval metrics computed from the logs. Secondary outcomes included hallucination rates and guideline citation behavior. Inter-rater reliability was assessed using intraclass correlation coefficients (ICCs). Results: In a paired within-model design that held each base model fixed, adding transparent guideline retrieval improved accuracy—significantly in the three weaker open-weight models (Mistral, Qwen3, and GPT-OSS) and directionally in the already-strong DeepSeek and GPT-5 bases. Because a pre-specified blinding check found that experts could still identify retrieval-augmented answers with 98.5% accuracy, we anchored causal interpretation on measures that do not depend on the human raters, ranked by their independence: deterministic, log-derived retrieval metrics first, and then an automated, condition-blind LLM judge, whose agreement with the experts (Spearman ρ = 0.81, 95.7% within-one agreement) establishes shared calibration rather than independence from their bias. Deterministically from the retrieval logs, citation groundedness rose from 0% to 51–89% and retrieval recall@5 was 92%. On the judge, content-level hallucination fell from 42% to 4% and accuracy rose by a pooled +0.64 points (95% CI 0.47–0.80); the accuracy gain persisted after adjustment for response length (+0.48, 95% CI 0.22–0.73), which retrieval shortened rather than lengthened. The accuracy gain was large for weaker base models and small or non-significant for already-strong ones, whereas the hallucination and auditability gains were consistent across all models. The human ratings reproduced the judge’s accuracy effect (+0.61, 95% CI 0.49–0.74), and GPT-5 run through the transparent pipeline showed no significant difference from the proprietary Custom GPT (judge accuracy 4.48 vs. 4.58). Conclusions: Guideline retrieval yields a reproducible, largely base-independent improvement in the safety and auditability of LLM answers to clinical guideline questions, with accuracy gains concentrated in weaker base models. Because retrieval-augmented answers are recognizable to experts, rigorous evaluation should rely on rater-independent measures, and residual hallucination continues to require human oversight. Full article

The continuous-review $(s,S)$ inventory system poses a fundamental analytical challenge: the cycle time and reorder undershoot are not independent when demand inter-arrival times are non-exponential, and scalar transform methods cannot resolve their joint distribution. We introduce a phase-tagged matrix functional $\mathrm{\Psi }(s,S;\xi ,\theta ,z)$, an $m\times m$ matrix that encodes, in a single closed-form object, the joint distribution of the reorder index, the cycle time, the reorder undershoot, and the operational phase of the inter-arrival process at the reorder epoch for any phase-type $\mathrm{PH}(\alpha ,S)$ inter-arrival distribution. The functional is derived via a finite discounted-visit recurrence with $O\left({\mathrm{\Delta }}^2\right)$ computational cost, where $\mathrm{\Delta }=S-s$ is the inventory cushion. From it we extract, in closed form, the expected demands per cycle, the mean cycle length, the undershoot distribution, and the long-run average cost rate. The classical scalar results of Sahin’s work in 1983 are recovered as the projection $\alpha \mathrm{\Psi }(s,S)e$. We prove a phase-coupling impossibility theorem: under PH renewal inter-arrivals, the phase at reorder is independent of any subsequent lead-time demand, delineating precisely where Markovian arrival process (MAP) inter-arrivals are needed. All closed forms are verified against $N=500,000$ independent Monte Carlo replenishment cycles on two instances with structurally different PH generators, with all relative errors below $0.5\%$. Full article

Objectives: Precise lymph node staging remains a cornerstone in the management of early-stage and locally advanced non-small cell lung cancer (NSCLC), directly influencing surgical planning and multimodal therapy. Despite the widespread use of 2-[18F]FDG PET/CT, occult nodal metastases frequently lead to unexpected upstaging after surgery, potentially affecting prognosis and therapeutic strategies. This study aimed to investigate whether radiomic features derived from preoperative PET/CT scans can predict nodal involvement in patients with early-stage lung cancer. Methods: A retrospective analysis was conducted on 124 patients with cT1N0 NSCLC who underwent 2-[18F]FDG PET/CT scans as part of the preoperative workup, followed by anatomical lung resection and systematic mediastinal lymph node dissection. Radiomic features were extracted from PET predictive of pathological nodal upstaging. Results: During the study period, 67 patients who underwent anatomical lung resection for early-stage lung cancer demonstrated unexpected nodal metastasis; a continuous series of 57 patients with the same clinical TMN was enrolled as a control group. Several radiomic parameters were significantly associated with nodal upstaging. According to variable importance (VIMP) analysis, metabolic tumor volume (MTV), total lesion glycolysis (TLG), run-length non-uniformity (RLNU), and gray-level non-uniformity (GLNU) emerged as the strongest predictors of lymph node involvement. Conclusions: Although the clinical utility of these findings remains to be validated, radiomic analysis of 2-[18F]FDG PET/CT imaging offers non-invasive biomarkers that may enhance the preoperative prediction of nodal involvement in early-stage NSCLC. Integrating radiomics into clinical workflows could improve surgical decision-making, refine patient selection, and reduce the incidence of unforeseen nodal upstaging. Full article

Single-image depth from defocus is limited by blur-radius ambiguity: two object distances on different sides of the focal plane can produce similar point spread functions (PSFs). We study single-shot dual-focus imaging (DFI) with a polarizer-free liquid crystal (LC) lens, where the ordinary-ray component remains unmodulated while the extraordinary-ray component is refocused. Unlike conventional multi-capture DFD, the proposed system records a dual-focus superposition in one exposure and selects the optical setting before network training using the ambiguity-interval length A and PSF correlation $C_r$. We show that DFI does not reduce ambiguity unconditionally: its benefit depends on the LC-lens power. A deblurring-based depth estimation network with a physics-calibrated Wiener bank translates the selected dual-focus cue into quantitative depth. On model-matched synthetic DFI data with an 8 m focus setting, the proposed configuration reduces RMS error from $0.227\pm 0.001$ m to $0.190\pm 0.001$ m (mean ± s.d. over four independent runs) relative to single-focus imaging. A 116-pair indoor prototype dataset provides feasibility evidence under the tested configuration, but is not used to claim broad physical generalization. Full article

This study addresses the safety and efficiency challenges of manual inspection in complex open-pit mining environments, where terrain steepness, limited coverage, and personnel exposure to hazards render conventional methods inadequate. We propose an integrated UAV inspection framework that combines 3D environmental modeling with a hierarchical optimization paradigm. The framework operates in three sequential stages. First, a high-fidelity 3D terrain model is constructed from point cloud data via skeletal feature extraction, which reduces computational complexity while preserving topographic structure. Second, an upper-layer Traveling Salesman Problem (TSP) solver determines the optimal inspection sequence across mandatory points (loading sites, dump sites, and crushing stations). Third, a lower-layer Chaotic Adaptive Population-based Grey Wolf Optimizer (CAP-GWO) refines the 3D path between consecutive TSP-ordered points, augmented by B-spline smoothing to ensure kinematic feasibility. Key inputs include: (i) raw LiDAR point cloud data of the mining site, (ii) facility coordinates and operational constraints (safety margins, maximum pitch angle, minimum turn radius), and (iii) UAV kinematic parameters. Outputs comprise a smooth, collision-free 3D trajectory with verified constraint satisfaction. Comparative experiments against eight metaheuristic algorithms (PSO, GA, ACO, BA, COA, GWO, SRA, SFOA) demonstrate that the proposed method reduces total path length by 15–20% on synthetic benchmark scenarios while maintaining zero constraint violations. Statistical validation via the Sign Test confirms the significance of these improvements (p < 0.05) across repeated independent trials. The framework is further validated on measured airborne LiDAR data of the Bingham Canyon open-pit copper mine (Utah, USA; USGS 3D Elevation Program), one of the largest operating open-pit mines in the world: on this real terrain, CAP-GWO achieves the best performance among the GWO-family algorithms, with a statistically significant 12.5% improvement over SRA (Wilcoxon p < 0.001) and 24% lower variance than the standard GWO, and all 210 experimental runs produce collision-free trajectories. Notably, the proposed hierarchical optimization framework achieves structural symmetry between the upper-layer sequencing task and the lower-layer path refinement task. This symmetric decomposition significantly reduces computational complexity while preserving solution quality, aligning with the principles of symmetry in engineering optimization. The framework offers a practical solution for autonomous, adaptive inspection scheduling in dynamic mining environments. Full article