Search Results (540)

Archaeological feature detection at complex, mixed-environment sites requires accurate, efficient methods for identifying subtle morphological signatures. This study presents a semi-automated remote sensing pipeline for the detection and delineation of archaeological earthworks at Cahokia Mounds (Illinois, USA), a major Mississippian urban center and UNESCO World Heritage Site. Three LiDAR datasets, two collected via UAV-mounted sensors and one from a piloted aircraft survey, were processed into Digital Terrain Models and transformed into Local Relief Models (LRM). K-means clustering was applied to segment the LRMs into feature classes, followed by contour bounding using the OpenCV library to outline mounds and borrow pits. Additionally, SAR-derived Local Incidence Angle (LIA) rasters from PALSAR-3 and Sentinel-1 were processed through angular deviation mapping to identify slope anomalies associated with archaeological features. Results across all five datasets demonstrate the complementary strengths of LiDAR and SAR: LiDAR excels at resolving elevation-defined features such as mound footprints, while LIA captures directional slope behavior that highlights mound edges, borrow pit rims, and linear features such as causeways. Comparative analysis of LiDAR acquisition frequencies reveals minimal differences in archaeological feature recovery between pulse settings, suggesting that sensor platform choice matters more than power-density tradeoffs for this application. Despite the need for human review to filter modern disturbances and natural false positives, the integrated workflow meaningfully accelerates prospection and reduces interpretive subjectivity. The methods are scalable, site-invariant, and work with open-access data, making them applicable to archaeological landscapes worldwide. Full article

Objectives: To assess the feasibility and reproducibility of predicting antimicrobial resistance (AMR) in Escherichia coli from MALDI-TOF mass spectrometry data using a standardized, open-source machine learning (ML) workflow, we systematically compared four ML algorithms, evaluated the impact of culture conditions, extract storage, and spectral preprocessing on model performance, and validated results through nested cross-validation with statistical significance testing. Methods: A total of 282 clinical E. coli isolates were analyzed. Two MALDI-TOF MS datasets were generated from freshly cultured extracts (T₁) and recultured isolates one year later (T₃), yielding 4468 spectra. A third dataset from the T₁ extracts stored at −20 °C for one year (T₂) was evaluated for spectral stability but excluded from primary modeling likely due to storage-induced degradation. Protein spectra (m/z 2000–15,000) were preprocessed using an in-house developed MALDI-TOF preprocessing pipeline (MTPP) comprising variance stabilization, Savitzky–Golay smoothing, SNIP baseline correction, TIC normalization, LOWESS alignment, and MAD-based peak detection (SNR ≥ 3), yielding 121 m/z features. Four classifiers—Random Forest (RF), Logistic Regression, Support Vector Machine, and Gradient Boosting—were trained to predict resistance to 11 antibiotics using nested cross-validation: outer GroupShuffleSplit (5-fold, isolate-level) for evaluation and inner GroupKFold for recursive feature elimination (RFECV) and hyperparameter tuning (RandomizedSearchCV). Classification thresholds were optimized via the precision–recall curve. Model performance was assessed using AUROC, AUPRC, F1-score, Matthews Correlation Coefficient (MCC), and bootstrap 95% confidence intervals (1000 replicates). Pairwise model comparisons were tested with McNemar’s chi-squared test. Results: Among the 12 antibiotics included in the analysis (meropenem excluded for absence of resistance), resistance prevalence ranged from 1.1% (colistin) to 59.9% (amoxicillin). Colistin was subsequently also excluded from ML modeling due to insufficient resistant isolates (n = 3), leaving 11 antibiotics for prediction. The best predictive performance was observed for ciprofloxacin (AUROC 0.76 [95% CI 0.74–0.77]; F1 0.54; MCC 0.38) and ceftazidime (AUROC 0.68 [0.65–0.71]; F1 0.36; MCC 0.29), using 13 and 37 RFECV-selected features, respectively. Amoxicillin achieved the highest F1-score (0.76), driven by high recall (0.98) but modest AUROC (0.58). No meaningful predictive signal was detected for amikacin, cefepime, or tigecycline (AUROC ≤ 0.57, F1 ≤ 0.17), attributable to extreme class imbalance, and no robust multi-peak resistance signature was detected in this dataset. McNemar’s test confirmed that RF significantly outperformed Logistic Regression for all antibiotics (p < 0.01), while Gradient Boosting performed comparably to RF for ciprofloxacin (p = 0.17) and ceftazidime (p = 0.28). Frozen extracts (T₂) produced lower spectral similarity and were excluded from model training; the aligned T₁₊₃ dataset yielded the most stable performance across metrics. Conclusions: Machine learning analysis of MALDI-TOF spectra enables reproducible AMR prediction for selected antibiotics in E. coli, with ciprofloxacin and ceftazidime showing the strongest signal. Nested isolate-level cross-validation, multi-model comparison with statistical testing, and open-source code provide a transparent, reproducible foundation for integrating ML-assisted MALDI-TOF analysis into diagnostic AMR surveillance. Extract storage at −20 °C degrades spectral quality and should be avoided in ML training workflows. Full article

As calibration errors have a direct impact on epipolar consistency, rectification accuracy, and metric 3D reconstruction performance, stereo camera calibration is a fundamental requirement for high-accuracy 3D modeling and reliable digital twin data acquisition. Because current calibration workflows (based on pairwise calibration methods) lack systematic data-quality checks mechanisms, there is a clear need for more robust data selection strategies. The novelty of the approach consists in the development of a new outlier-aware stereo calibration algorithm (OutAw) that introduces a unified multi-stage approach that integrates hard geometric selection, candidate subset generation, multi-criterion ranking, bootstrap stability analysis, and triangulation assessment into a comprehensive and systematic calibration framework. Unlike conventional approaches, OutAw (through its mechanism of detecting and rejecting inconsistent pairs) redefines the calibration strategy from arbitrary to criterion-based data selection. Also, the proposed algorithm is compared with BSC (a baseline OpenCV all-pairs calibration algorithm) and InterFil (an intermediate filtered variant) using 49 stereo pairs (at 1280 × 720 resolution) captured using a planar checkerboard. OutAw algorithm achieved (using only nine image pairs) superior results (epipolar error 0.5119 px, stereo RMS 0.7666 px) to the BSC ones (epipolar error 1.3687 px, stereo RMS 1.9385 px), representing statistically significant improvements (60.5%, respectively 62.3%). OutAw geometric consistency was validated by triangulation-based metrics (square-length standard deviation 0.1140 mm and square absolute error 0.1097 mm). Contamination analysis revealed that as the outlier rate increases, the calibration process degrades progressively. Also, the results obtained highlight that geometric quality-driven image selection is critical for achieving a reliable stereo calibration for DT applications. Full article

Traditional acupoint localization methods rely heavily on manual operation, resulting in high subjectivity and limited accuracy. To improve the precision and stability of acupoint detection, this study integrates machine vision technology with in situ projection to achieve automated recognition and real-time visualization of human acupoints. First, an automatic calibration method based on image processing is proposed for back acupoints. Spinal features are extracted from the blue channel, enhanced using adaptive histogram equalization, and processed through region of interest extraction, minimum-threshold binarization, and morphological operations. Key spinal curve points are then fitted using Bézier functions. Canny edge detection is used to extract the human silhouette, locate the acromion, and derive the pixel scale of the “cun” measurement, enabling coordinate computation for 141 back acupoints. In the deep learning component, an improved YOLOv8-Pose model is developed for acupoint localization. Unlike existing methods that use local attention or the original Object Keypoint Similarity (OKS) loss, we introduce two innovations: a non-local attention module for global dependency modeling, and a novel Efficient Object Keypoint Similarity (EOKS) loss function that incorporates geometric constraints—namely, width, height, and center distance—in addition to Euclidean distance. A non-local attention mechanism is incorporated into the backbone to enhance global feature extraction, and the EOKS loss function is designed to improve spatiogeometric regression accuracy. An inference mechanism is further introduced to derive the remaining acupoints from 49 detected keypoints; experiments demonstrate that the improved model achieves 95.0% detection accuracy, outperforming the baseline by 2.62%, with an inference time of 14.5 ms. Finally, an in situ projection platform is constructed, combining camera calibration, four-point proportional scaling, and an OpenCV 4.5.4-based interactive interface. The system supports real-time translation, rotation, and scaling, enabling accurate projection of detected acupoints onto the human body. Full article

The growing worldwide prevalence of Diabetes Mellitus highlights the urgent need for effective early detection methods to enable prompt intervention. This study develops a machine learning-based decision-support prototype for predicting diabetes risk using health metrics from the DiaHealth dataset, a recently published Bangladeshi open-source dataset for Type 2 diabetes prediction. Five supervised learning algorithms were evaluated: Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Decision Tree (DT), and Random Forest (RF). Models were assessed across three stages: before feature scaling, after standardisation, and following hyperparameter optimisation via GridSearchCV, using accuracy, precision, recall, and F1-score as evaluation metrics. LR and SVM showed marked improvements after standardisation, consistent with their sensitivity to feature magnitude, whilst tree-based approaches such as DT and RF remained largely unchanged. KNN displayed minimal sensitivity to scaling, which is discussed in relation to the feature distributions of the dataset. Following hyperparameter tuning, RF achieved the highest accuracy of 95%, outperforming all other models. RF predictions were interpreted using Local Interpretable Model-agnostic Explanations (LIME) to promote transparency in model decision-making. The best-performing model was subsequently deployed as an interactive web-based prototype application using Streamlit, providing real-time prediction outputs. These findings demonstrate how preprocessing choices and hyperparameter tuning can differentially affect algorithm performance and illustrate the potential of combining explainable AI with practical deployment for diabetes risk assessment in a research context. Full article

Angelica dahurica (A. dahurica) is an important medicinal plant in China; however, its production is affected by viral infections, leading to reduced yields and quality. In this study, we identified a novel carlavirus, tentatively named Angelica carlavirus virus (AnCV), in the leaves of A. dahurica exhibiting mosaic and leaf crinkling symptoms. Notably, the complete genome of AnCV was 8562 nt long and contained six open reading frames, with a genomic organization typical of the genus Carlavirus. AnCV exhibited 44.5–57.8% nucleotide identity at the whole-genome level with known members of the genus Carlavirus. In the polymerase gene and coat protein regions, the highest nucleotide and amino acid identities were 59.4–60.0% and 46.5–55.8%, respectively, which were below the species demarcation criteria established by the International Committee on Taxonomy of Viruses for the genus Carlavirus. Importantly, 10 AnCV isolates clustered within subgroup I of the genus Carlavirus, forming a relatively distinct branch. Moreover, 119 of the 280 A. dahurica samples were positive for AnCV (detection rate of 42.58%). Our study revealed that AnCV is a novel member of the genus Carlavirus that infects A. dahurica, providing a theoretical basis for the monitoring and control of viral diseases in A. dahurica. Full article

Concrete cracks are ubiquitous in practical engineering structures and continuously affect structural safety and durability. Crack images provide important visual evidence of damage evolution; however, crack images alone are insufficient to determine the residual load-bearing capacity of concrete members. Although the development of deep learning algorithms has significantly improved the automatic detection of concrete surface cracks, most existing methods remain limited to the extraction of crack geometric features and lack a direct connection with mechanical performance. To explore the relationship between image-based crack geometry and mechanical response, this study combines U-Net-based crack segmentation, OpenCV-based crack geometry extraction, and phase-field fracture simulation to establish a preliminary visual–mechanical framework for plain concrete beams. In this framework, surface crack images are first segmented using a U-Net model, and crack length, average width, and propagation path are extracted from the predicted binary masks. The extracted crack length is then used as the primary variable to match the observed crack state with the phase-field crack evolution sequence. Once the corresponding simulation stage is identified, the associated load level and residual load-bearing capacity can be obtained from the simulated load–crack mouth opening displacement (Load–CMOD) response. Through a mixed-mode I–II fracture test, the crack geometric features extracted by deep learning are compared with the phase-field simulation results. The results show that the error in crack length is within 2.5%. Meanwhile, the relative error between the simulated peak load and the experimental value was 1.57%, which preliminarily verified the correlation between image-based crack information and the load-bearing capacity of plain concrete beams. The method is further applied to a Mode I fracture test without recorded load-bearing capacity data. By mapping the crack length identified from the image, namely 36.89 mm, to the phase-field evolution sequence, the load-bearing capacity of the member at this stage is estimated to be 74.4% of the peak load. The results indicate that the crack geometry extracted from images can be correlated with phase-field crack evolution, thereby supporting preliminary residual load-bearing capacity assessment of plain concrete beams. However, the proposed framework should be regarded as a case-level feasibility study rather than a general structural assessment method. Before broader engineering application, further validation using synchronized crack image sequences, crack mouth opening displacement (CMOD) measurements, and load records is required. Full article

In this study we describe the refinement process from the first to the second phase of a teaching–learning sequence development and implementation. The TLS comprises several experimental activities that aim to support understanding of Control of Variables Strategy (CVS) reasoning in the context of floating/sinking and properties of magnets. The research was carried out during a science laboratory course in a department of early childhood education. The participants numbered 67 in the first phase of the survey and 45 pre-service early childhood teachers (referred to as student teachers) in the second phase. The analysis is theoretically grounded in Pickering’s model of scientific practice, as adapted in science education, which provides the analytical framework for identifying and categorizing refinement changes. The results showed that the refinements are differentiated from each other according to the factors that guide them. Specifically, the three refinement changes guided by the educational factor were local-guided, i.e., related to a specific activity dealing with the student teachers’ educational needs, and the other two, also driven by the scientific factor, were holistic-open refinements, i.e., related to a set of activities adjusting the TLS to the new scientific trends. These findings contribute to the literature on Teaching-Learning Sequence development by illustrating how theoretically grounded analysis can make refinement processes more explicit and analytically interpretable. Full article

Efficient feed management is essential in aquaculture, especially for bottom-feeding species such as shrimp that require feed delivery at the tank bottom. Most commercial automated feeders are designed for surface-feeding fish and are unsuitable for benthic organisms, leading to feed waste and uneven distribution. This study developed and evaluated an automatic bottom feeder capable of dispensing sinking pellets directly to the substrate. The system integrated a 3D-printed auger for precise feed metering and a closed-type centrifugal impeller positioned at the water surface to achieve radial dispersion of feed. An Arduino Uno microcontroller operated the impeller speed (285.98–586.85 rpm), feed mass (95.23–285.68 g), and dispersion time (2–8 s). A Box–Behnken response surface methodology was used to analyze the influence of these parameters on the mean radius spread of feed, supported by image-based uniformity assessment using OpenCV. Results identified impeller speed as the most significant factor (p = 0.010), with optimal dispersion observed at moderate speeds and longer spread durations. The system demonstrated reliable mechanical performance and precise control, providing a novel, programmable solution for uniform feed delivery in shrimp aquaculture and a promising foundation for scalable, automated bottom-feeding technologies. Full article

Fog-dependent Tillandsia landbeckii in the hyper-arid Atacama Desert lacks the red-edge reflectance pattern that supports vegetation monitoring, motivating shortwave infrared (SWIR) approaches. We evaluated a newly developed UAV-borne multispectral SWIR camera system for estimating plant water status and additional plant functional traits (fresh and dry biomass, and N uptake) from four spectral bands (1100, 1200, 1510, and 1650 nm) across 20 destructively sampled plots. Of five traits tested, only canopy water content (CWC) retained statistically robust spectral associations after multiple-testing correction, with most significant predictors concentrated in the 1200–1510 nm wavelength region. A physically interpretable predictor, the mean spectral slope between 1200 and 1510 nm, yielded conditional cross-validated $R_{\mathrm{cv}}^2=0.51$ (${\mathrm{RMSE}}_{\mathrm{cv}}\approx 170$ g m⁻²), though fully selection-corrected estimates were substantially lower ($R_{\mathrm{cv}}^2=0.10$–$0.20$), reflecting feature-selection instability at the given sample size. The absence of robust biomass- and nitrogen-related signals is physically interpretable given the species’ atypical surface optics. While expanded sampling and independent validation remain necessary to establish transferable performance estimates, these results demonstrate that SWIR-based water-status retrieval is feasible for this spectrally challenging species, opening a pathway toward functional monitoring of fog-dependent desert ecosystems. Full article

This paper presents an adaptive nonlinear control and state observation framework for energy management in standalone photovoltaic (PV) systems integrated with battery energy storage. A unified nonlinear dynamic model is developed to describe the interactions between the PV generator, the DC/DC buck converter, and the lithium-ion battery. Based on this model, a multi-mode control strategy is designed to ensure efficient and safe operation under varying environmental and loading conditions. The proposed scheme incorporates maximum power point tracking (MPPT) to maximize photovoltaic energy extraction, along with constant current (CC) and constant voltage (CV) charging modes to guarantee battery safety and longevity. To address uncertainties and unmeasured states, an adaptive nonlinear observer is developed for real-time estimation of the battery open-circuit voltage and state of charge. The observer design is supported by Lyapunov-based stability analysis, ensuring boundedness and convergence of the estimation error in the presence of modeling uncertainties and external disturbances. An energy management algorithm is further introduced to coordinate the transition between operating modes according to the estimated system states and battery constraints. The effectiveness and robustness of the proposed control and observation strategy are validated through detailed simulations in MATLAB/Simulink under varying solar irradiance conditions. The results demonstrate accurate maximum power tracking, reliable state estimation, and safe battery charging performance, highlighting the potential of the proposed approach for advanced autonomous PV–battery systems. Full article

This paper proposes an intelligent stress detection method based on convolutional neural networks and the DeepFace framework, addressing the challenges of increasingly prominent global mental health issues and the limitations of traditional psychological services in terms of early warning latency and coverage. A three-level cascaded strategy combining RetinaFace, MTCNN, and OpenCV is first employed for face detection and localization, and facial expression features are extracted via the DeepFace framework. By integrating Russell’s valence–arousal model with Lazarus’s cognitive appraisal theory, an emotion–stress mapping rule is constructed to convert seven-category emotion probability distributions into 1–5 scale stress values. The method employs a cloud–edge collaborative flow, with feature extraction performed at the edge and original images promptly destroyed to mitigate privacy risks. Experiments on public expression datasets indicate that the method achieves above 99% face detection accuracy, 84.99% emotion recognition accuracy, and 86.09% stress assessment consistency grounded in the emotion–stress mapping rule, with an average response time per frame of approximately 200 ms. Based on 233 multi-scenario surveys, some respondents show limited stress self-awareness, suggesting traditional self-reporting may have blind spots, and thus this method serves as a useful supplement. Full article

Background/Objectives: Active surveillance (AS) is recommended for men with low-risk prostate cancer to minimize overtreatment while monitoring for disease progression. However, current surveillance strategies rely heavily on repeat biopsies, which are invasive and associated with morbidity. MyProstateScore 2.0—Active Surveillance (MPS2-AS) is a urine-based biomarker test developed to predict progression to Grade Group ≥ 2 (GG ≥ 2) and Grade Group ≥ 3 (GG ≥ 3) prostate cancers in men on AS. The objective of this study was to analytically validate the reproducibility and robustness of MPS2-AS analyte detection and risk score calculation across key laboratory variables. Methods: Analytical precision was evaluated using pooled urine specimens processed using the MPS2-AS laboratory workflow. Eight pooled urine samples were tested in a within-laboratory design across five days, with two runs per day, and two replicates per run. Additional reproducibility studies assessed variability across three QuantStudio™ 12K Flex Real-Time PCR Systems and three OpenArray™ chip lots. Ten RNA biomarkers were quantified by RT-PCR and used to calculate the MPS2-AS GG1-2 and GG1-3 risk scores. Variance components were estimated using hierarchical ANOVA. Results: The MPS2-AS analyte measurements demonstrated high precision across within-laboratory testing, with standard deviations ranging from 0.00 to 0.60 and coefficients of variation (%CV) from 0.00 to 4.01%. The reproducibility across qPCR instruments and OpenArray chip lots showed similar robustness, with analyte %CVs of ≤4.57% and ≤4.10%, respectively. These stable analyte measurements translated to reproducible model outputs, with %CV ≤ 10.69% for the GG1-2 risk score and ≤7.20% for the GG1-3 risk score across all tested conditions. No systematic bias was observed between runs, days, instruments, or reagent lots. Conclusions: MPS2-AS demonstrates strong analytical precision and reproducibility for quantifying urinary biomarkers and generating GG1-2 and GG1-3 risk scores. These results support the reliability of MPS2-AS for clinical laboratory implementation and its use as a non-invasive tool to inform biopsy decisions in men with Grade Group 1 prostate cancer undergoing active surveillance. Full article

Background and Objectives: Open-angle glaucoma subtypes share a structural phenotype but differ in pathophysiology: pseudoexfoliative glaucoma (PXG) involves vascular endothelial dysfunction associated with deposition of exfoliative material, whereas normal-tension glaucoma (NTG) reflects primary vascular dysregulation in the absence of elevated intraocular pressure. We characterized subtype-specific OCT angiography (OCTA) profiles obtained from a 3 × 3 mm macular scan and evaluated their discriminatory power for pairwise subtype classification. Materials and Methods: This was a single-center, cross-sectional study of 304 eyes: 198 glaucomatous eyes—primary open-angle glaucoma (POAG, glaucoma simplex in our clinical nomenclature), n = 102; PXG (glaucoma capsulare), n = 68; NTG (glaucoma sine tensio), n = 28—and 106 healthy controls. The Cirrus HD-OCT 5000 AngioPlex 3 × 3 mm OCTA protocol was used to assess vessel density (VD), perfusion density, foveal avascular zone (FAZ) morphology, ganglion cell complex (GCC), and retinal nerve fiber layer (RNFL) thickness. Analyses included Kruskal–Wallis tests with Bonferroni post hoc correction, ROC analysis with DeLong comparison of combined versus structural-only models, multivariate regression, and an exploratory XGBoost classifier with SHAP-based interpretation. Results: VD Inner and Perfusion Inner were lower in PXG (16.37 ± 3.33%; 0.31 ± 0.05) than in POAG (18.73 ± 3.41%; 0.34 ± 0.05; both p < 0.001); Perfusion Inner was also lower than in NTG (p < 0.05). FAZ Area was largest in NTG (0.27 ± 0.11 mm²) and greater than in PXG (0.19 ± 0.08; p < 0.01); FAZ Circularity differed across subtypes (p < 0.001). Combined OCTA–structural models outperformed structural-only models for POAG vs. PXG (DeLong p = 0.002) and for PXG vs. NTG (AUC = 0.770; p = 0.010). Sector-resolved Spearman analysis revealed subtype-specific coupling: in NTG, VD Inner and Perfusion Inner correlated with the inferior RNFL (r = 0.53 and r = 0.52; both p < 0.01); in PXG, coupling shifted nasally (r = 0.41 and r = 0.46; both p < 0.001). The exploratory XGBoost classifier separated glaucoma from controls with an internal cross-validated AUC of 0.975 ± 0.008 (5-fold CV; not externally validated); FAZ Circularity (mean |SHAP| = 0.418) and FAZ Area (0.411) were the top inter-subtype features, supported by case-level SHAP. RNFL avg and average GCC independently predicted MD across subtypes; in PXG, Perfusion Inner also predicted MD (β = −32.78; p = 0.032). Conclusions: In this single-center, cross-sectional cohort, OCTA revealed subtype-associated macular microvascular profiles that are complementary to structural OCT. Reduced vessel and perfusion density characterized PXG, whereas FAZ enlargement and reduced circularity distinguished NTG and PXG. Vascular–structural coupling was nasal-predominant in PXG and inferior-predominant in NTG. Combined multimodal models outperformed structural-only approaches. Macular perfusion additionally predicted MD in PXG. The XGBoost/SHAP analysis is exploratory; prospective and externally validated studies are required before clinical deployment. Full article

To meet the demand for real-time and accurate diamond particle size measurement in industrial scenarios—where traditional image processing methods lack robustness in complex environments and existing deep learning models struggle to balance accuracy and efficiency—this paper proposes an integrated framework for dynamic segmentation and morphological analysis of diamond particles based on video streams. A fully automated data acquisition system consisting of a high-precision motion stage, an industrial camera, and an optical microscope is first constructed to capture dynamic particle images. Based on YOLOv8n-seg, a lightweight SPR-YOLOv8 instance segmentation model is then developed with three key improvements: a Large Separable Kernel Attention (LSKA) mechanism is introduced into the SPPF module to enhance feature discriminability; the RepBlock module is adopted in the neck network to improve feature fusion efficiency through structural re-parameterization; and a P2 small-object detection head is introduced while large-object detection branches are removed, enabling the model to focus on tiny, densely distributed particles. Finally, a contour-based geometric analysis method is proposed for particle size estimation based on segmentation results. Experimental results show that the proposed model achieves an mAP@0.9 of 0.861 while maintaining a low parameter count (0.97 M) and a high inference speed of 500 FPS. Compared with the conventional OpenCV-based method (CADPS), the proposed DPSCA framework reduces the mean absolute percentage error in particle size measurement by over 70%, while also demonstrating strong accuracy and stability in consecutive-frame tracking. Overall, this study provides a practical and efficient automated inspection solution for online quality control in superhard material manufacturing, and supplementary cross-scale experiments further demonstrate promising robustness on diamond particles beyond the primary 180–250 μm range. Full article