Search Results (1,980)

Accurate faulty feeder identification is essential for the safe operation of small-current grounded distribution networks under single-phase-to-ground faults, yet existing transient-based methods often rely on preset data windows and insufficiently exploit the collective consistency of healthy feeders. This study proposes a training-free method coupling adaptive transient window selection with robust multi-feeder graph consistency analysis. Its novelty is the joint determination of which post-fault interval should be retained and which feeder deviates from the healthy feeder group. Zero-sequence current variations are obtained using one-cycle-earlier pre-fault reference waveforms, and an effective analysis window is selected within a preset maximum acquisition interval according to the cumulative transient energy ratio. The normalized current segments are then mapped to a weighted similarity graph, and a fused anomaly index based on average connection strength and median edge weight is used for feeder selection. A 110/10 kV six-feeder MATLAB/Simulink system with 180 fault cases was evaluated. With ${\eta }_0=0.98$, the method achieved 98.89% accuracy with an average selected analysis-window length of 29.62 ms; accuracies under 10 and 5 dB noise were 96.17% and 91.78%, respectively. It also outperformed Haar wavelet energy and Wasserstein distance baselines in accuracy and normalized separation. These results support its effectiveness as a physically interpretable, training-free feeder identification framework. Full article

Stereo vision measurement in industrial robot environments fails on thin structures such as needles and gripper tips, because both traditional and deep-learning-based stereo matching produce boundary errors there. We identify and verify experimentally that ground-truth disparity inflation in public datasets is a systematic error source that pushes networks toward biased boundary estimates. From this, we classify the resulting measurement deviations into four categories: edge errors, intra-segment expansion, small-object loss, and uniform-region distortion. We then propose a geometric-constraint-driven measurement refinement method that corrects each category in turn. Since every correction step is geometric rather than learned, the method is unaffected by ground-truth inflation, a property that conventional post-processing filters do not offer. A GUM-based uncertainty propagation analysis of the measurement model $D=f·B/d$ shows that disparity uncertainty dominates the depth uncertainty budget when f and B are exactly known. Experiments on KITTI 2015, Middlebury, and a custom UE4 synthetic industrial dataset (100 stereo pairs) with nine stereo baselines (seven deep learning and two traditional) show that, on inflation-free ground truth, the refinement imposes a near-zero systematic penalty on deep learning output while clearly improving traditional methods. On KITTI, the predictable metric shift confirms that the method is unaffected by LiDAR ground-truth inflation. On a real industrial robot scene, the refined disparity recovers the gripper tip and needle that the baseline LEAStereo loses. These results position geometric-constraint-driven refinement as an effective, training-data-independent complement to end-to-end stereo matching for precision industrial measurement, within the tested scenes and methods. Full article

Unmanned Aerial Vehicle-borne Ground-Penetrating Radar (UAV-GPR) has been used for ice thickness monitoring in lakes and rivers due to its non-contact measurement, high resolution, and operational flexibility. Existing algorithms can extract ice layer boundaries by tracking continuous bottom reflections in GPR images. However, they fail when the radar signal lacks a clear bottom reflection—a common condition in ice layers containing unfrozen water—and manual interpretation remains time-consuming. To address this limitation, this paper builds a freshwater ice GPR dataset covering both fully frozen and unfrozen water-bearing zones, and proposes a method for ice thickness identification based on the DeepLabv3+ neural network. The model performs pixel-level binary classification, labeling each pixel as ice layer or background, and generates a segmentation mask that constrains the subsequent thickness calculation to valid ice regions only. Field validation against drilling measurements demonstrates that the model achieves Intersection over Union (IoU) of 97.12% and an F1-score of 98.54% for ice layer identification, with a relative error in ice thickness measurement below 3% based on five borehole measurements. Field tests in two reservoirs across Tibet and Jilin, China, demonstrate that the proposed method can accurately characterize the distribution and thickness of the ice layer while effectively eliminating the interference of unfrozen water zones. The results demonstrate that the proposed method can provide automated, accurate ice thickness estimates for UAV-GPR surveys of freshwater ice. Full article

Distributed Acoustic Sensing (DAS) has revolutionized ambient noise interferometry, yet the reliability of retrieved empirical Green’s functions (EGFs) remains highly sensitive to non-diffuse noise fields and anthropogenic transients. In complex or noisy environments, global metrics frequently misclassify signal convergence due to out-of-band environmental noise, scattered coda, and non-stationary directional transients. Using both 30 min and 4 h passive recordings, this study presents a physics-guided quality control framework for evaluating interferometric diagnostics. Specifically, this study employs the signal-to-trailing noise ratio (STRN), signal-to-precursory noise ratio (SPNR), and spectral signal-to-noise ratio (SSNR) within the surface-wave arrival window $t=x/v$. Global phase metrics remain heavily suppressed $(\overline{x}\approx 0.05)$ regardless of stacking duration, whereas the surface-windowed SPNR exhibits an extraordinary statistical shift (p < 0.001), reaching 0.870 ± 0.106 at 30 min and 0.967 ± 0.034 at 4 h. We implement one-to-one correspondence between surface-windowed indicator values and fundamental-mode Rayleigh wave dispersion sharpness. In severely noise-contaminated segments, unwindowed global metrics yield distorted dispersion ridges with severe energy leakage, but the surface-wave window results in an increase in the SPNR above 0.70, fully reconstructing continuous dispersion trajectories (250–500 m/s). Grounded in these results, we formalize a standardized four-step quality control workflow (from velocity windowing to metric calculation, automation, and data output) and outline tailored adaptation guidelines for urban, mountainous, and industrial DAS deployments. This framework provides an automated, physically sound protocol that eliminates manual selection, optimizes computational efficiency, and ensures reliable dispersion extraction for passive DAS imaging. Full article

Multi-frame fusion is standard for converting frame-by-frame license plate character detections into stable, reliable readings. Character Time-Series Matching (CTM), a leading approach, associates characters across frames using the Hungarian algorithm with a fixed Euclidean distance threshold and a translation-only motion model, reporting 96.7% accuracy on the UFPR-ALPR dataset. In this work, we demonstrate that this high performance is protocol-dependent: when ground-truth static plate crops and pre-segmented tracks are used, CTM performs strongly. However, in real-world scenarios involving moving cameras (such as drone-mounted cameras, helmet-mounted cameras, and mobile platforms) where inter-frame geometry changes dynamically, baseline multi-frame association frameworks that combine fixed spatial gates with unconstrained translation propagation fail. In these cases, temporal fusion provides no benefit and degrades plate recognition performance below the single-frame baseline. Indeed, under its own Intersection over Union (IoU) tracker, this literature method correctly reads only 15.1% of plates in traffic videos recorded with a real moving camera (86 human-verified tracks). To address this vulnerability, we propose Geo-CTM (Geometry-Constrained CTM), an association pipeline integrating height-scaled adaptive matching gates, inter-frame similarity estimation via Random Sample Consensus (RANSAC), transform-guided character coasting, and co-occurrence-constrained duplicate track elimination. Systematic motion-model ablation demonstrates that while the complete association pipeline provides the primary foundation for robustness (raising mean accuracy from 85.40% to over 91.5%), estimating a similarity transform (91.82%) delivers the most physically grounded and identifiable representation on planar plates without estimation degeneration. While our method performs comparably to CTM on ideal data when using the same detector and detections, it minimizes performance loss under geometric distortion conditions where CTM is inadequate. For instance, a statistically significant improvement is achieved under a 0 → 60° perspective change; in real traffic videos, with the tracker held fixed so that the fusion layer is the only variable, performance rises from 15.1% to 26.7% under the IoU tracker of the original system and from 16.3% to 29.1% under ByteTrack ($+11.6$ and $+12.8$ points; exact McNemar $p=0.021$ and $p=0.013$), whereas changing the tracker alone while holding the fusion layer fixed moves accuracy by only 1–2 points and is not statistically significant. Finally, our error taxonomy analysis demonstrates that on the undistorted benchmark all residual errors correspond to zero-evidence cases beyond the reach of decision-level fusion, while under dynamic perspective distortion errors are dominated by association misalignment, highlighting the specific development areas that future performance improvements must target. Full article

When the power-frequency short-circuit current flows through the armor rods segment on an overhead ground wire (OGW), the OGW at the segment may experience fracture failure due to high temperatures. Consequently, it is necessary to optimize the structural configuration of the armor rods segment. Based on the structural characteristics of the conventional armor rods segment, this paper proposes a stepped-type armor rods segment structure. First, an electromagnetic–thermal coupling simulation model for both types of armor rods segment is constructed, in which the conductor length is determined by the boundary conditions of both the electromagnetic field and the thermal field. The current density distribution and transient temperature distribution under power-frequency short-circuit current are analyzed using the simulation model. Subsequently, based on the simulation results, an evaluation method for the mechanical performance of the OGW considering non-uniform temperature distribution is proposed. This method is employed to compare the high-temperature mechanical properties of the OGW at the two types of ends. Finally, a transient temperature rise experiment is designed to validate the accuracy of the simulation model. The research results show that the simulation model has sufficient accuracy, with an error of no more than 6%. Compared with the conventional armor rods segment, the stepped-type structure effectively avoids the concentration of high-temperature zones. Under identical conditions, the mechanical load-bearing capacity of the OGW at the stepped-type end is higher than that at the conventional end, which can help prevent high-temperature fracture of the OGW to a certain extent. Full article

In open-field agriculture, timely human activity recognition (HAR) is critical for anticipating worker actions and enabling proactive human–robot collaboration. This study developed an offline early HAR framework based on pre-segmented activity sequences. Long Short-Term Memory (LSTM) networks were used in conjunction with wearable inertial sensors mounted on the chest, cervical region, lumbar region, and right and left wrists. Accelerometer, gyroscope, and magnetometer signals were collected from 20 participants during an outdoor agricultural material-handling task, with a mobile ground robot serving as the receiving platform for the crate. Each activity was represented by cumulative prefixes from 30% to 100% of its duration. At the 30% observation ratio, the model achieved a macro-F1 score of 0.9314, compared with 0.9481 for the full sequence, corresponding to an absolute difference of 0.0167. Shapley Additive Explanations (SHAP) were also used to identify the body locations, sensor modalities, and signal channels that contributed most. The early decisions were mainly supported by sensors placed on the trunk, with wrist sensors providing complementary information, particularly for standing classification. Multimodal inertial information was also important. The most influential inputs were mainly gyroscope and magnetometer channels from the chest, cervical region, and lumbar region. In conclusion, the high performance at early observation ratios highlights the potential to support more adaptive and better-coordinated robot-assistance strategies. Full article

Disaster response often requires evidence from several sources, including satellite imagery, social media, news reports, audio and video recordings, and event metadata. Existing disaster datasets, however, usually focus on one source, a limited set of modalities or a single task. This separation makes it difficult to evaluate models that must combine regional observations with ground-level evidence for the same disaster context. We introduce DisasterScope, an event-centric multimodal dataset and benchmark for disaster-type recognition, severity assessment and evidence retrieval. DisasterScope organizes satellite observations, social images, synchronized audio–video segments, observation text, report passages, and provenance metadata around canonical events. Its primary benchmark contains 12,159 fixed event-context bundles from 24 events and nine disaster classes. The dynamic-media inventory includes 273 source videos, 628 curated audio–video segments, 9.02 h of material and 5473 temporal windows. Labels are harmonized through source-label inheritance, taxonomy mapping, teacher assistance and partition-specific human review. Validation and test annotations were reviewed in full, while training annotations were sampled for review and accepted through a threshold gate. The benchmark evaluates full-input and unavailable-view conditions on the same bundle identities, allowing direct measurement of how each view affects a model without changing the evaluated sample population. DisasterScope therefore provides a traceable setting for studying multimodal disaster assessment across different disasters, evidence sources and input-availability conditions. Full article

Background/Objectives: Accurate delineation of the optic disc and optic cup in retinal fundus photographs is a prerequisite for automated glaucoma screening. While encoder–decoder segmentation models have advanced considerably, the contribution of upstream preprocessing to segmentation accuracy, and the stability of that contribution across repeated training runs, remain insufficiently characterized. Methods: Five preprocessing pipelines, baseline, Contrast Limited Adaptive Histogram Equalization (CLAHE), Region of Interest (ROI) cropping, ROI+CLAHE, and CLAHE with heavy augmentation, were benchmarked under a fixed EfficientUNet++ model with an EfficientNet-B7 encoder on three publicly available fundus datasets (REFUGE, ORIGA, and Drishti-GS). Every configuration was retrained under three independent random seeds (42, 15, and 89) to assess run-to-run variability. Seed-level standard deviations accompany every reported mean and define the confidence limit on each ranking. Results: On REFUGE, CLAHE with augmentation (Config 5) achieved the strongest mean Dice (disc $0.9523\pm 0.0017$; cup $0.8348\pm 0.0018$). On ORIGA, all five configurations clustered within 0.0067 disc Dice; ROI+CLAHE (Config 4) was marginally ahead on disc ($0.9681\pm 0.0002$) and augmentation led on the cup ($0.8873\pm 0.0024$). On Drishti-GS, all five configurations converged successfully once optimizer and loss settings were corrected; the near-total failures seen in earlier single-run experiments reflected a configuration problem, not the small (81-image) training set. Conclusions: CLAHE applied to full-resolution images is the single most consistently beneficial preprocessing choice across all three datasets. ROI+CLAHE showed a small, initialization-stable advantage on ORIGA, but ROI crop centres were derived from ground-truth centroids, an oracle localization setting, and these results should not be interpreted as achievable by a fully automated pipeline. Data augmentation showed a consistent reduction in initialization sensitivity on small datasets and may be beneficial as a default strategy. Full article

This study examines whether perceived interactivity is associated with tourists’ continuance intention toward generative AI travel assistants through two parallel post-use evaluations—perceived usefulness and flow experience—while accounting for individual technology-related dispositions. Drawing on the Stimulus–Organism–Response (S–O–R) framework, the model conceptualizes perceived usefulness as a cognitive–instrumental evaluation and flow experience as an experiential–absorptive evaluation, while examining personal innovativeness in information technology and technology anxiety as focal moderators, selected on theoretical grounds, of the interactivity–flow and interactivity–usefulness associations, respectively. Data were obtained through a purposive online survey of 853 tourists with prior experience using generative AI for travel-related purposes, and the proposed relationships were tested using partial least squares structural equation modeling (PLS-SEM). The findings show that perceived interactivity is positively associated with continuance intention, perceived usefulness, and flow experience. Both perceived usefulness and flow experience are positively associated with continuance intention and exhibit significant indirect associations between perceived interactivity and continuance intention when estimated simultaneously. The indirect association through perceived usefulness is numerically larger than that through flow experience, while the remaining direct association is consistent with the two evaluations providing a complementary but non-exhaustive account of continuance. Personal innovativeness is positively associated with flow experience, and the positive interactivity–flow association is estimated to be stronger at higher levels of personal innovativeness. Technology anxiety is negatively associated with perceived usefulness, and the positive interactivity–usefulness association is estimated to be weaker at higher levels of technology anxiety. Although statistically significant, both interaction magnitudes are small and therefore provide limited rather than strong evidence of association-specific heterogeneity across tourists. The model demonstrates moderate explanatory performance and positive but limited benchmark-relative out-of-sample predictive relevance. The study contributes to research on generative AI-enabled tourism by showing that perceived interactivity is concurrently associated with distinct and nonredundant instrumental and experiential forms of post-use value, each of which retains a separate association with continuance intention. This dual pattern is particularly relevant to travel decision-making, in which tourists frequently navigate complex and interdependent choices under experiential uncertainty while remaining engaged in an evolving planning process. Within the S–O–R framework, the study specifies a context-bound dual-evaluation account of generative AI post-adoption rather than claiming that the inclusion of additional mediators or moderators constitutes a general extension of the framework. The findings offer general practical considerations for designing generative AI travel assistants that combine responsive interaction with functional value and engaging user experiences. However, disposition-specific design and segmentation implications should be treated as tentative given the small interaction magnitudes. Full article

Variable-rate pesticide spraying and fertilizer application require coordinated operation of heterogeneous agricultural machines, particularly in irregular fields where task demands, vehicle mobility, payload capacity, energy consumption, and resupply requirements vary spatially. However, most existing studies optimize aerial spraying or ground fertilization separately and do not jointly consider prescription-map demands, air–ground synchronization, pesticide-drift risk, and agricultural vehicle constraints. This study formulates collaborative UAV spraying and UGV fertilization as a multi-objective mixed-integer nonlinear programming problem with three objectives: minimizing system makespan, weighted energy consumption, and pesticide-drift penalty. A prescription-map-guided bi-level planning framework is proposed. At the upper level, the problem-specific TNSAOO solver determines UAV and UGV task sequences and collaborative resupply-point activation. At the lower level, adaptive Theta* and row-constrained Hybrid A* generate UAV spraying and UGV fertilization trajectories, respectively, while prescription-dependent application commands are assigned along active operation segments and a time-window mechanism detects and corrects residual air–ground conflicts. The framework was evaluated using 30 real farmland boundaries and 90 randomized prescription scenarios. Mean geometric coverage rates reached 98.82% for UAV spraying and 98.95% for UGV fertilization, while the mean prescription-compliance errors were 6.21% and 2.13%, respectively. In addition, 96.7% of the batch runs contained no more than one detected air–ground conflict, with a mean corrective waiting time of 1.07 s. Compared with traditional independent operation, collaborative planning reduced mean system makespan by 9.32%, weighted energy consumption by 7.21%, modeled drift penalty by 3.62%, and total path length by 5.79%. In the multi-objective comparison, TNSAOO obtained a mean hypervolume of 0.597 and a mean inverted generational distance of 0.375, showing competitive Pareto-search performance relative to established comparison algorithms, particularly NSGA-II. Additional terrain and drift sensitivity analyses produced systematic changes in energy, completion time, and modeled drift risk under controlled parameter perturbations. These findings demonstrate the simulation-based feasibility of jointly planning heterogeneous variable-rate spraying and fertilization under a shared prescription map. Physical field experiments remain necessary to validate spray deposition, fertilizer-distribution uniformity, terrain effects, and model calibration under environmental uncertainty. Full article

Inland-vessel speed reductions targeted at priority segments can reallocate a normalized burden proxy across a fixed corridor. We develop a route-integrity speed-planning model that accepts a plan only when it attains the priority target, preserves route-wide improvement, and caps the largest non-priority-segment increase. The kinematics distinguish speed over ground, along-route current, and speed through water; without voyage-matched current or speed-through-water observations, the cases use zero-current analytical references. We evaluate two AIS-derived cases: the 27-point Case A as a sparse stress test and Case B as the primary spatial case after WGS84 and ESA WorldCover alignment checks. Under the equal-time constraint, an unprotected 2% target induces local deterioration. With a 0.5% non-priority cap, Case B supports a certified maximum target of 0.999%; a 2% late-arrival allowance raises the certified maximum target to 11.912%. Direct search over 0.1-kn commands finds a feasible plan for Case B; however, none of 1000 sampled ±0.2-kn execution-error profiles remains feasible. The framework identifies model-internal proxy transfer rather than measured energy, emissions, or exposure effects. Full article

Overlapping cells in plant suspension-culture microscopy pose a particular challenge, for instance, segmentation because a single pixel may belong to more than one cell. Most standard instance-segmentation methods are not designed for this setting and tend to treat overlapping objects as mutually exclusive regions. We instead represent each cell as an independent full-cell instance and introduce MC-SlotNet, an architecture that separates competitive object-slot feature assignment from mask decoding. This allows multiple predicted masks to occupy the same image region. We further introduce a mask-level multiplicity-consistency loss that encourages the predicted number of masks covering a pixel to agree with the underlying cell occupancy. We evaluate MC-SlotNet on a newly annotated dataset of 53 Siraitia grosvenorii suspension-culture micrographs containing 4131 full-cell instances acquired at 4×–40× magnification. Using grouped five-fold cross-validation and an overlap-preserving evaluation protocol, we compare the method with Mask R-CNN, SOLOv2, and Mask2Former. MC-SlotNet achieves the best performance on AP50 (0.800), mAP$50:95$ (0.565), F150 (0.842), all-ground-truth Dice (0.761), AJI+ (0.754), overlap-region Dice (0.705), and overlap-instance recall (0.828). Its AP75 (0.649) is comparable to Mask2Former’s (0.651). MC-SlotNet also has the lowest inference time among the evaluated methods, at 1.113 s/image. These results indicate that decoding full-cell masks independently, rather than enforcing an exclusive partition of image pixels, is well-suited to instance segmentation in plant suspension-culture microscopy images with substantial cell overlap. Full article

Ultra-deep circular shafts are increasingly deployed in congested urban environments, yet their structural performance in highly sensitive soft clay remains susceptible to diaphragm wall panel joints, structural imperfections, and complex interactions with adjacent structures. This study presents a high-fidelity three-dimensional (3D) numerical investigation into the excavation behavior of an ultra-deep circular shaft with a diameter of 30 m and an excavation depth of 56.3 m in Shanghai soft clay by synthesizing high-resolution field monitoring with advanced finite element modeling. The numerical framework was established in PLAXIS 3D utilizing the Hardening Soil model with small-strain stiffness (HSS), explicitly incorporating an equivalent structural stiffness reduction scheme (0.8 vertically and 0.5 circumferentially) to capture panel segmentation, joint compliance, and concrete cracking. The reduced-stiffness model successfully reproduces the measured deep-seated bulging profiles and internal force distributions with high fidelity. The findings reveal exceptional deformation control capabilities of the circular geometry, yielding a maximum lateral wall deflection of merely 9.1 mm (0.016%H_e), which is significantly smaller than the normalized deformation ratio of 0.3%H_e observed in five analogous rectangular excavations in Shanghai. The numerical results indicate that circumferential compression governs the overall load transfer behavior, while vertical bending response remains relatively limited. Furthermore, a pronounced circumferential anisotropy in wall deformation is governed by asymmetric boundary conditions, where localized Metro Jet System (MJS) ground improvement significantly restrain movements, whereas the non-grouted area experience peak deflections. Crucially, interaction with the adjacent external diaphragm walls of ancillary structures induces a complex 3D stress redistribution rather than a beneficial shielding effect, amplifying the peak shaft wall displacement by nearly 62.8% (from 4.73 mm to 7.70 mm). These insights underscore the criticality of integrating small-strain soil mechanics, equivalent structural degradation, and adjacent structural boundaries into predictive design protocols for ultra-deep circular retaining systems. Full article

Corrosion-severity grading of grounding-grid connectors from optical images supports proactive power-infrastructure maintenance. Existing approaches rely on single-time-point, manually thresholded hue–saturation–value (HSV) metrics and static multi-criteria decision-making (MCDM) frameworks that cannot capture corrosion dynamics. In this paper we present a pipeline that (1) defines a four-class corrosion grade from an HSV area fraction (S_corr) measured on RGBA optical images, and validates those labels against a baseline-referenced CIEDE2000 metric zero-referenced to each connector’s as-received appearance; (2) generates 240 color-prior-constrained procedural synthetic images from 53 real images across six connector types; (3) fine-tunes a ResNet-18 to estimate corrosion coverage continuously, deriving the reported severity class from that estimate rather than predicting it directly; and (4) fits power-law kinetics C(t) = k·tⁿ to the S_corr time series, propagates bootstrap uncertainty into a Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) framework, and reports kinetics-aware rankings as rank probabilities. The label validation quantifies two limitations of single-threshold HSV grading: a material-color offset that scores an unexposed copper connector at S_corr = 0.442, and insensitivity to achromatic corrosion products covering roughly 80% of the aluminum and galvanized-steel surface. Ablation experiments replicated over five random seeds show that neither contribution claimed from a single run survives replication: synthetic augmentation changes macro-F1 by +0.050 (p = 0.46) under the adopted checkpoint-selection rule and by −0.059 (p = 0.43) under the rule used in the original experiments, and the monotonicity-consistency loss by −0.011 (p = 0.87) and +0.001 (p = 0.99) respectively; the previously reported single-run values of 0.208 and 0.494 are draws from opposite tails of the same seed distributions (0.403 ± 0.140 and 0.344 ± 0.073). The one formulation that improves significantly is the continuous one adopted here, which raises Spearman agreement with the independent metric from 0.316 ± 0.150 to 0.698 ± 0.108 (p = 0.005). Measured against controls, a classifier that never sees the image reaches macro-F1 = 0.425 and, after Holm–Bonferroni correction, no deep configuration is distinguishable from it; none exceeds a one-dimensional linear rule on S_corr (0.664); and under leave-one-material-out cross-validation the network does not improve on S_corr used directly as a predictor (ρ = +0.627 against +0.744, paired p = 0.14). Time-resolved energy-dispersive X-ray spectroscopy (EDS) provides a partial chemical consistency check, with welding at ρ = 0.82 (raw p = 0.023), but no material survives Holm correction across the six tested. A U-Net segmentation head supervised only by synthesis-derived masks attains Dice = 0.85 in-domain and collapses to a 0.033 output range on real images, 5% of the HSV metric’s range; the photometric-stability advantage previously claimed for it is an artifact of that collapse and is withdrawn. Kinetics-aware MCDM with propagated uncertainty resolves 9 of 15 pairwise orderings, placing stainless steel above welding at 30 chamber days with probability 1.000 and reversing the static ranking. The pipeline, code and fixed data split are fully reproducible (random seed 42). Full article