Search Results (1,295)

A collision-free path is not sufficient for visibility-dependent mobile robot tasks: a moving obstacle can block the camera–target line of sight and cause inspection or visual-servoing failure even when the robot remains physically safe. Maintaining visual contact with targets is therefore important in Industry 4.0 environments, yet visibility-preserving maneuvers can conflict with navigation progress and collision avoidance. This work presents the Visibility-Informed Safety and Target Awareness framework with control barrier function filtering and occlusion-evasive local replanning (VISTA-CBF+ELR). The architecture combines visibility-risk planning, target-bearing control, an ELR supervisor, and a CBF quadratic program that keeps collision constraints hard while relaxing field-of-view and occlusion requirements through slack. Counterproductive interventions are limited through persistence, benefit–cost and feasibility gates, progress protection, bounded dwell, recovery, and cooldown. In locked factory simulations, redesigned VISTA achieved 67% and 73% strict-goal success under clean and nominal sensing, whereas Visibility-CEM-2D achieved 87% and 86% but with lower clearance. In matched Gazebo trials, strict success was 19/30 for redesigned VISTA, 26/30 without ELR, and 16/30 for Nav2 Smac+MPPI; zero-clearance collisions were 8/30, 3/30, and 14/30, with no difference surviving multiplicity correction. A separate CEM stress test sustained 6.875 Hz optimization, missed 26.31% of 100 ms deadlines, and held commands on 31.35% of ticks. The results demonstrate repair of the ELR pathology and conditional visibility-risk reduction while exposing safety–visibility trade-offs, transfer limitations, and real-time constraints. Full article

Solar ultraviolet (UV) radiation is a major contributor to skin damage, making the regular use of broad-spectrum sunscreens essential for effective photoprotection. However, the long-term efficacy of sunscreen formulations is limited by the photoinstability of some organic UV filters, particularly avobenzone. Licuri oil (Syagrus coronata) is a naturally derived Brazilian palm oil that represents a promising alternative to medium-chain triglycerides (MCTs) as an oily core for polymeric nanocapsules. Therefore, this study aimed to develop Eudragit^® RS 100-based cationic nanocapsules using licuri oil for avobenzone encapsulation and to incorporate them into hyaluronic acid (HA) or xanthan gum (XG) hydrogels. The nanocapsules ranged from 125 to 158 nm, with a polydispersity index (PDI) of less than 0.2, positive zeta potential (+11 to +13 mV), and encapsulation efficiency above 98%. After 48 h of UVA exposure, the nanocapsule formulations retained 51% and 52% of their initial avobenzone content for the licuri oil- and MCT-based systems, respectively, compared with only 10% for free avobenzone, indicating a marked improvement in photostability following nanoencapsulation. Additionally, licuri oil nanocapsules exhibited increased antioxidant activity compared to MCT-based nanocapsules in both DPPH and ABTS assays. The nanocapsule suspensions were classified as non- to slightly irritating in the Hen’s Egg Test–Chorioallantoic Membrane (HET-CAM) assay. After incorporation into the hydrogels, the resulting formulations exhibited pseudoplastic and thixotropic behavior, with HA-based hydrogels providing higher UV absorption than XG-based hydrogels. Furthermore, no transdermal permeation of avobenzone was observed from either the HA- or XG-based hydrogel formulations, and the HA hydrogel containing licuri oil nanocapsules showed higher stratum corneum retention. These findings demonstrate the potential of licuri oil as an effective alternative to MCTs as the oily core of polymeric nanocapsules for topical formulations containing encapsulated avobenzone. Full article

Symbolic and rule-based learning offers a transparent, sample-efficient alternative to the statistical paradigm that dominates contemporary machine learning. Whereas large language models and deep networks approximate target functions from massive corpora without exposing the rules they rely on, many practical problems instead call for compact, human-readable concept definitions learned from only a handful of examples. In this paper, we introduce the Concept Tree Learner (CTL), an incremental and interpretable symbolic framework that induces logical concepts over binary strings from minimal labeled data. CTL represents knowledge as a rooted tree of logical predicates; it extends this tree incrementally as each labeled example is processed and, after observing the data, distills the simplest rule set consistent with all examples through a set-cover filtering step, in accordance with Occam’s Razor. We give formal definitions for the tree structure, its construction and pruning operations, and the rule-selection objective, and we analyze the worst-case time and space complexity of the procedure. A prototype implementation, operating over a small set of atomic binary-string predicates whose numeric arguments scale dynamically with the input length, is evaluated across 29 concept-learning tasks of varying complexity. CTL recovers a consistent concept for every task—the intended one on 26 of the 29 datasets, and an equally consistent alternative on the three whose training set does not uniquely determine it—typically converging before all training examples are exhausted, and produces fully interpretable rule sets. On the same atomic vocabulary, it generalizes substantially better to unseen strings than both a classical entropy-based decision tree and the RIPPER (Repeated Incremental Pruning to Produce Error Reduction) rule learner ($95.5\%$ mean accuracy across all tasks—and $100\%$ on the 26 tasks whose training set uniquely determines the target—versus $72.1\%$ and $77.2\%$ respectively), while using fewer and shorter rules. We position CTL within the literature on symbolic machine learning, inductive logic programming, and decision-tree induction, discuss the current limitations of the prototype—its restriction to noise-free binary input, its batch (sorted) training regime, and its fixed predicate vocabulary—and outline concrete directions for extending it toward a self-expanding, hierarchical concept-learning system. Full article

To explore an alternative detection approach for the pesticide imidacloprid (IMI), this study repurposed the quinoline-benzimidazole fluorescent probe DQBM-B—previously developed for Co²⁺ recognition—and investigated its detection performance and interaction mechanism toward IMI in the aggregated state. The optimal working conditions of the probe were determined by optimizing key detection parameters, and the sensing performance and matrix compatibility were evaluated through selectivity tests and proof-of-concept spiked cucumber extract analysis. The DQBM-B aggregates interact with IMI synergistically through intermolecular hydrogen bonding and π–π stacking, which enrich IMI at the aggregate surface to create a local enrichment layer. The observed fluorescence quenching arises from the synergistic contribution of static quenching (due to ground-state complex formation) and the inner filter effect (IFE). Under the optimal conditions, the system exhibited a detection limit of 0.75 μmol L⁻¹ for IMI with favorable anti-interference ability. The matrix effect evaluation in cucumber extract demonstrated good recovery and precision, demonstrating the feasibility of the aggregation-regulated IFE strategy in complex food matrices. This study expands the application scope of the DQBM-B probe from metal-ion sensing to pesticide detection and provides a metal-free, aggregation-regulated strategy for the fluorescence detection of neonicotinoid pesticides. Full article

Although the Pacific oyster (Crassostrea gigas) is a premium marine protein source, its neuroprotective peptidome remains largely uncharacterized. This study established an integrated in silico and in vitro pipeline to discover acetylcholinesterase (AChE)-targeting peptides with cellular AChE-regulating and neuroprotective peptides from simulated gastrointestinal digests of oyster. Peptidomic profiling identified 18,292 sequences, which were filtered down to seven candidates predicted to have favorable blood–brain barrier (BBB) permeability and to be non-toxic and non-allergenic (VPYPR, VPVHF, HHTF, PVHF, GPKPW, HWF, and KYW) via multi-step virtual screening. In cellular assays, simulated H₂O₂ injury (500 μM) reduced PC12 cell viability to 47.53 ± 4.53%. Compared with the model group, pretreatment with the three most potent candidates—HHTF, VPYPR, and VPVHF (200 μM)—significantly rescued injured cells, restoring cell viability to 88.31 ± 7.83%, 85.12 ± 3.35%, and 82.00 ± 3.47%, respectively (p < 0.05). These peptides effectively fortified cellular antioxidant defenses by increasing glutathione (GSH) levels to 24.24, 30.11, and 26.83 nmol/mg protein (from 20.22 nmol/mg protein in the model group) and superoxide dismutase (SOD) activity to 151.41, 153.97, and 151.96 U/mg protein (from 119.33 U/mg protein), while suppressing malondialdehyde (MDA) accumulation to 0.088, 0.064, and 0.086 nmol/mg protein (from 0.193 nmol/mg protein). Crucially, the peptides alleviated cholinergic dysfunction by normalizing the H₂O₂-induced elevation of intracellular AChE activity (11.39 nmol/min/mg protein) down to 7.02, 6.22, and 7.14 nmol/min/mg protein, respectively. Specifically, VPYPR (200 μM) restored AChE activity to a level (6.22 nmol/min/mg protein) that was not significantly different from that in the normal control group (p > 0.05). Molecular dynamics (MD) simulations (100 ns) and molecular mechanics Poisson–Boltzmann surface area (MM-PBSA) calculations identified VPYPR as the leading candidate with a remarkably low binding free energy of −49.74 ± 3.58 kcal/mol. This study demonstrates that oyster gastrointestinal digests are valuable reservoirs of multi-target neuroprotective ingredients and provides an efficient strategy for marine bioactive peptide discovery. Full article

Why do the laws represented by the Standard Model and general relativity have their observed forms, and why do their particular symmetry groups and numerical inputs occur? The reconstruction program proposes that these structures arise from a premetric selection layer logically prior to dynamical evolution. This article develops the laws side of that program as a conditional physical reconstruction framework with a theorem-level mathematical core. It is neither a replacement for established post-read-out physics nor a completed theory of quantum gravity. The Indefinite Reconstruction Stability Principle (IRSP) is formulated as an existence condition: once an observable quotient and a class of admissible continuations have been declared, a candidate physical identity cannot depend on an unread representative or on an arbitrary terminal refinement. A structure that fails this requirement may be an intermediate mathematical configuration, but it is not a reconstruction-independent physical identity. This persistence is not temporal immortality: unstable particles and other transient systems remain admissible when their identities and changes are unambiguously readable. An exact descent theorem formalizes representative independence. A single neutral parent, understood as a common premetric archetype rather than an additional particle, then organizes the conditional read-out bridges. Primitive complement readability, compact scalar return, and smooth holonomy yield codimension-two loop structure and a local $\mathrm{U}\left(1\right)$ connection. A continuum stability implementation of IRSP forces real-root hyperbolicity, while a separate full-dimensional reciprocal-response bridge supplies a nondegenerate quadratic principal form and hence Lorentzian signature with one temporal direction. Conformal descent of the reconstructed curvature two-form independently fixes the total dimension to four. On the selected $(3+1)$-dimensional platform, gauge invariance and nativeness yield a source-free Abelian Maxwell law, while a readable metric scale yields Einstein–Hilbert dynamics at leading order. A separate conditional $3+2$ internal read-out selects $S\left(\mathrm{U}\right(3)\times \mathrm{U}(2\left)\right)$, equivalently, the Standard Model gauge group with maximal ${\mathbb{Z}}_6$ quotient. These results are conditional on explicitly audited bridges. Reconstruction may consequently offer quantum-gravity programs, including string theory, an additional persistence-and-read-out filter on candidate microscopic phases or vacua; applying that filter to solve the landscape problem remains an open program rather than a result established here. Full article

Active-power ceilings create a control challenge in droop-controlled grid-forming converter networks because the electrical response is faster than the measurements and outer control. Projected and projection-free power limiting use filtered active power in the outer power–frequency channel and therefore cannot act directly on the first electrical power peak. This paper proposes constraint-activated projection-free control, which coordinates a shaped projection-free multiplier with a bounded resistance term in the capacitor-voltage reference driven by instantaneous terminal power. A general full-order dynamic model describes the converters, controllers, and network without tying the formulation to a particular benchmark. Local well-posedness is established, and the proposed controller is shown to preserve the constrained projection-free equilibrium and active-branch Jacobian, allowing the same full-order stability assessment. Across ten tested scenarios with unchanged controller parameters, the proposed controller reduces peak power exceedance by 38.7–55.4% and accumulated excess energy by 34.1–62.2%. The corresponding DC-buffer requirement decreases without activating the independent current limiter, which isolates the source-side power constraint from AC overcurrent. A network-level study demonstrates sequential transitions between one and two constrained sources while the remaining converter supplies the feasible power imbalance. Full-order stability verification, component studies, and parameter sweeps establish the role and useful range of each controller path. Full article

Parkinson’s disease (PD) is associated with motor impairment and altered cortical dynamics. Although resting-state electroencephalography (EEG) has been widely studied for PD classification, walking EEG remains comparatively underexplored despite its relevance to gait dysfunction. This study developed an EEG-only, leakage-free framework for participant-level classification of PD and healthy controls (HCs) using resting- and walking-state EEG data from the OpenNeuro ds007526 dataset. Following selection, preprocessing, and quality control, 132 participants were retained, including 109 individuals with PD and 23 HCs. EEG recordings were resampled to 128 Hz, filtered between 0.5 and 40 Hz, screened for artifacts, harmonized across conditions, and segmented into overlapping 6 s windows. Classical machine-learning (ML) models used participant-level engineered features, whereas deep-learning (DL) models classified pooled resting-state and walking windows. Extra Trees achieved the highest balanced accuracy (0.71) among the ML, while ShallowConvNet-Lite was the strongest standard DL baseline (0.82). The proposed NeuroAtten-PD model achieved an accuracy of 0.86, a balanced accuracy of 0.84, an F1-score of 0.91, an ROC-AUC of 0.90, and a sensitivity of 0.88. After Holm correction, participant-level paired comparisons indicated significant differences from DeepConvNet-Lite and EEGNet-Lite. These findings demonstrate the feasibility of participant-level PD classification using a pooled collection of resting-state and walking EEG windows. With validation in larger, independent, and more balanced clinical cohorts, the proposed framework could support objective EEG-based decision support and provide a foundation for portable or wearable systems for longitudinal monitoring of PD-related cortical changes in clinical and home-based settings. Full article

Federated, semi-supervised novelty detection is well suited for intrusion detection on resource-constrained Internet of Things (IoT) nodes: each device learns a model of benign traffic, shares only summary statistics, and does not transmit raw traffic samples. A previously published cross-layer federated detector, Baseline K-Means, showed that periodically merging worker statistics through a coordinator raises the detection rate, but it also exhibited a systematic side effect: after every merge, the precision decays, and the false-positive rate (FPR) climbs because the coordinator recomputes its threshold from streaming distances filtered by the closest observed anomaly, so tightens after every merge, flagging progressively more benign traffic; the threshold was also hand-tuned. We present AF-BKM, an Adaptive Federated Baseline K-Means that repairs the federated mechanism with two label-free, statistics-only enhancements, denoted as E1 and E2: (i) an adaptive decision threshold read from the benign Mahalanobis-distance distribution, requiring no manual percentile search and no attack labels (E1), and (ii) a robust, benignly anchored aggregation that blends worker means under quality weighting and outlier-worker filtering and recalibrates the threshold on a trusted benign anchor to a stable, anchor-referenced false-positive level, which a target-FPR rule can make operator-selectable instead of tightening it toward the nearest anomaly (E2). With MinMax scaling fit only on benign baseline data and non-IID federated streams on NSL-KDD, UNSW-NB15 and the N-BaIoT corpus of real traffic from commercial IoT devices, AF-BKM removes the merge-induced precision decay (the first-to-last-epoch precision change improves from $-0.134$ to $-0.002$ on NSL-KDD, from $-0.121$ to $-0.014$ on UNSW-NB15, and from $-0.170$ to $-0.009$ on N-BaIoT) and reduces the mean FPR by 30–$64\%$, depending on the dataset; all central improvements are significant across 10 seeds (Wilcoxon $p=0.002$, large effect sizes). AF-BKM preserves recall on NSL-KDD and N-BaIoT and, on the harder UNSW-NB15, exposes an explicit precision–recall trade-off through a benign target-FPR knob. In fp32, the deployed model serializes to $5.5$–52 KB, a packet is classified in 11–27 µs on a desktop CPU, and each merge round uploads a $d+3$-value summary (160–472 B) $94.6$–$98.3\%$ smaller than the same summary extended with the covariance upper triangle. A robustness study covering selected faulty-worker updates, contamination of the commissioning anchor, and detector-level white-box evasion reports the measured degradation patterns: fabricated threshold candidates have no direct path to the threshold, although a fabricated mean still reaches it indirectly through the blended centroid, and the anchor-referenced false-positive level remains stable under percent-level anchor contamination, while recall sensitivity is dataset-dependent and the evasion budget tracks the benign–attack margin of each dataset. We frame the contribution with a focused taxonomy that identifies merge-induced precision decay under non-IID workers as an open gap. Code is released for reproducibility. Full article

Caspase-1 is a crucial inflammatory cysteine protease that facilitates the maturation of pro-inflammatory cytokines such as interleukin-1β and interleukin-18, making it a significant therapeutic target for inflammatory diseases. However, existing caspase-1 inhibitors often face challenges like toxicity and suboptimal drug-like properties, underscoring the need for new inhibitors. This study employed an integrated computational strategy, combining quantitative structure–activity relationship (QSAR) modeling and application of this validated model to a large natural product database followed by molecular docking, rigorous binding free energy analysis and extended molecular dynamics simulations. Initially, a dataset of 185 caspase-1 inhibitors with experimentally reported pKi values (ranging from 4.05 to 9.24) was used to construct a QSAR model using Partial Least Squares (PLS) regression. The PLS-based QSAR model was developed with 18 descriptors out of 5799 calculated descriptors for each compound and 10 latent variables, demonstrating strong statistical performance with R² values of 0.870 and 0.838 for the training and test sets, respectively, and leave-one-out cross-validation coefficient Q²_LOO and 5-fold cross-validation (Q²_5-fold) values of 0.819 and 0.814, respectively. Y-randomization tests further confirmed the model’s robustness, as the randomized models exhibited significantly lower statistical parameters than the original model. The validated QSAR model was applied to 276,518 natural products in the LOTUS database. Subsequent molecular docking, Molecular Mechanics/General Born Surface Area (MM/GBSA) scoring, and Pan-Assay INterference Compounds (PAINS) and Chemical Frequent Hitter (ChemFH) filtering identified 14 candidate compounds, which were further evaluated using 300 ns molecular dynamics simulations. Among these, four natural products (LTS0162325, LTS0221286, LTS0016840, and LTS0070407) showed the most stable binding behavior and maintained persistent interactions with key catalytic and substrate-binding residues of caspase-1 in a mimicked physiological condition. Overall, this study highlights natural diterpenoids and coumarin glycosides as promising scaffolds for caspase-1 inhibition and demonstrates that integrating QSAR modeling with structure-based approaches provides an efficient strategy for discovering potential anti-inflammatory drug candidates. Full article

Camera-reconstructed 3D scenes often require offline visual cleanup before inspection, presentation, or reuse as renderable virtual-scene assets. Representative applications include removing temporary furniture, parked vehicles, equipment, signage, and other distracting or obsolete objects from reconstructed indoor and outdoor environments. Such editing requires not only accurate target localization across viewpoints but also plausible recovery of the previously occluded background. Existing methods often depend on manually specified masks or category-restricted detectors, while projection-based pipelines independently inpaint multiple views and subsequently refine the 3D representation, potentially introducing cross-view appearance and geometry inconsistencies. We present OVR-GS (Open-Vocabulary Removal in Gaussian Splatting), an instruction-driven object-removal framework for pre-trained 3D Gaussian Splatting (3DGS) scenes. Given a free-form instruction, a language parser generates target-oriented queries and a textual background-completion condition. Grounding DINO and the Segment Anything Model (SAM) produce multi-view candidate masks, which are filtered using Contrastive Language–Image Pre-training (CLIP). The proposed Semantic-Aware Gaussian Selector (SAGS) aggregates rendering-contribution-weighted mask evidence, groups spatially coherent candidates, and identifies the target Gaussian subset through rendered-cluster semantic verification. After removal, new Gaussians are initialized from boundary-adjacent primitives and interior samples and optimized locally using Score Distillation Sampling (SDS), while the original background remains fixed. On IMFine, SPIn-NeRF, and Inpaint360GS, OVR-GS achieves peak signal-to-noise ratio (PSNR) values of 19.78, 17.82, and 24.62 dB and Fréchet inception distance (FID) values of 142.30, 148.60, and 34.80, respectively. The results demonstrate the effectiveness of localized Gaussian optimization for instruction-driven cleanup of reconstructed environments before visual inspection, presentation, or reuse as renderable virtual-scene assets. Full article

Background/Objectives: Accurate verification of single-isocenter multiple-target (SIMT) stereotactic radiosurgery is challenging owing to the complexity of multi-lesion delivery and the sensitivity of stereotactic dose gradients. We aimed to evaluate the efficacy of electronic portal imaging device (EPID)-based multileaf collimator (MLC) transmission and dosimetric leaf gap (DLG) correction for improving portal-dose prediction agreement in SIMT stereotactic radiosurgery (SRS) and to propose an exploratory target count-based institutional action level. Methods: This retrospective analysis included 112 consecutive patients treated with HyperArc™-based SRS (1–13 targets). Treatment plans were calculated using the Acuros XB algorithm for 6 MV flattening filter-free (FFF) beams. Two sets of MLC parameters were compared for portal-dose image prediction (PDIP): (1) standard parameters measured using an ion chamber; (2) EPID-derived corrected parameters. Portal-dose accuracy was evaluated using gamma index analysis with a 95% pass rate threshold. Results: The uncorrected method showed a strong negative correlation between target number and gamma passing rates, with complex plans (≥9 targets) dropping as low as 70%. EPID-based correction substantially improved dose agreement, yielding consistent passing rates above 98%, regardless of target number. Although uncorrected parameters remained within tolerance for plans with one to two targets, accuracy declined markedly starting at three targets. Conclusions: Our findings indicate that the EPID-based correction of MLC transmission and DLG mitigated cumulative modeling discrepancies in complex SIMT SRS. Thus, we propose the exploratory institutional observation of three targets, beyond which EPID-based correction may be considered to achieve optimal portal-dose prediction agreement in HyperArc-based stereotactic radiosurgery. Full article

Micro- and nanoplastics (MNPs) are increasingly recognized as emerging contaminants in drinking water, yet quantitative data remain limited due to analytical challenges. In this study, the occurrence, morphology, and polymer composition of MNPs were investigated in commercial drinking water from Saxony-Anhalt, Germany, including ultrapure water, tap water, and five bottled mineral waters. A dual-stage cascade filtration system combining silicon filters (5 µm pore size) and aluminum oxide membranes (90 nm pore size) was employed for size-selective particle separation. Retained particles were characterized using optical microscopy, scanning electron microscopy (SEM), and Raman spectroscopy. Tap water exhibited the highest concentration of microparticles, whereas ultrapure water showed minimal microparticle contamination. Nanoparticles were detected in all samples, including ultrapure water, with mean particle sizes below 1 µm, highlighting the pervasive nature of nanoscale plastic contamination. Raman analysis identified polyethylene terephthalate (PET) as the most frequently detected synthetic polymer, together with polypropylene (PP), polystyrene (PS), and poly(methyl methacrylate) (PMMA). The results demonstrate that no drinking water source tested in this study was free of detectable MNPs and underline the need for standardized analytical approaches capable of reliably detecting MNPs in drinking water. Full article

Accelerometer-based tilt angle estimation is widely used in engineering monitoring, yet random noise and outliers degrade its accuracy. Integer-order Kalman filters suppress noise, but their Markovian model assumes that the current state alone is sufficient to predict the next, neglecting the influence of earlier states on slowly varying processes. Fractional-order Kalman filters incorporate historical states into the prediction. However, the conventional formulation shares a single transition matrix between the state and covariance predictions, underestimating the prediction uncertainty, while the memory mechanism propagates gross errors across iterations. To overcome these limitations, this paper proposes a decoupled fractional-order Kalman filter (DFKF). The method assigns independent transition matrices to the state and covariance predictions, where a scaling coefficient inflates the predicted covariance to lower the prediction weight and strengthen reliance on measurements. A front-end gross-error pre-elimination strategy combining second-order differencing with adaptive peak detection is further introduced to block outlier propagation at the source before it enters the memory mechanism. Simulations under varying noise levels and gross-error conditions show that DFKF achieves a mean RMSE (Root Mean Square Error) of 0.147°, representing reductions of 17.4% and 6.4% over KF (0.178°) and FKF (0.157°), respectively, and a mean MaxAE (Maximum Absolute Error) of 0.548°, outperforming KF and FKF by 24.5% and 10.3%. Under gross-error conditions, DFKF converges in 0.012 s on average, approximately 2.8 times faster than KF and FKF, and the pre-elimination strategy restores accuracy to near-error-free levels. Full article

Field-road cracks in high-standard farmland are often slender, low-contrast, and surrounded by complex textures that cause U-Net-based models to misclassify aggregates, tire marks, shadows, and repair edges as cracks. To reduce these false positives, this study develops a task-oriented context-guided hard-negative background suppression network (CGHN-Net) based on Squeeze-and-Excitation U-Net (SE-U-Net). The context-guided skip gate (CGSG), a same-resolution adaptation of additive attention gating, uses already upsampled decoder features as semantic guides to filter encoder skip features at all three scales. Hard-negative background suppression loss (HNBS Loss), a background-restricted hard-example mining objective, further targets elevated-probability responses within ground-truth background regions. The dataset comprised 2235 vehicle-acquired grayscale pavement images from independent sessions and mutually exclusive road sections: 1684 for training, 464 for validation, and 87 for testing. Across three random seeds, CGHN-Net achieved Dice, IoU, precision, recall, and FP area ratio values of 0.8682 ± 0.0055, 0.7791 ± 0.0102, 0.8780 ± 0.0161, 0.8734 ± 0.0266, and 0.0042 ± 0.0008, respectively. Against U-Net, Attention U-Net, UNet++, DeepLabV3+, SegFormer-B0, and BGCrack, it achieved the highest Dice, IoU, and recall, indicating the strongest overall overlap and crack recovery. Sequence-aware paired analysis against UNet++ preserved contiguous acquisition order through block lengths of 3, 5, and 10 images, and all block-bootstrap confidence intervals excluded zero. On 100 held-out crack-free images, CGHN-Net also achieved the lowest post-processed image-level false-alarm rate and FP area ratio among the included models. Additional three-seed validation on the independently acquired public CrackForest Dataset (CFD), with the selected models retrained on mutually exclusive CFD partitions, showed that CGHN-Net achieved Dice, IoU, and recall of 0.6679 ± 0.0162, 0.5028 ± 0.0182, and 0.9486 ± 0.0085, respectively, exceeding UNet++ and BGCrack in overlap and crack recovery. The results support the task-oriented combination of same-resolution skip filtering and background-restricted hard-example mining for suppressing texture-induced false responses, while the CFD experiment is interpreted as independent public-dataset retraining rather than zero-shot transfer. Full article