Search Results (6,723)

Mobile laser scanning (MLS) has become an effective technique for deformation monitoring in subway shield tunnels. Among various deformation characteristics, segment dislocation is an important indicator of tunnel structural health because it reflects the relative deformation between adjacent segments and may affect the mechanical behavior and waterproof performance of segmental joints. However, existing MLS-based methods for dislocation detection still suffer from facility interference, inaccurate seam localization, and limited automation in quantitative analysis. To address these challenges, this study proposes an automated method for shield tunnel segment dislocation detection based on MLS point cloud processing. The proposed framework consists of three main steps. First, a point cloud filtering strategy integrating offset features and semantic segmentation is developed to remove facility-related noise while preserving tunnel wall information. Second, a tunnel segment segmentation method combining bolt hole extraction and moving template matching is introduced to achieve accurate localization of both horizontal and longitudinal seams, where bolt holes are identified using normal vector and distance constraints. Finally, automated segment dislocation analysis is performed based on the filtering and segmentation results. Experimental results demonstrate that the proposed filtering method improves accuracy by 8.7% and 5.6% compared with conventional ellipse fitting and cylinder fitting methods, respectively. Using manually interpreted reference values derived from the same MLS dataset as the evaluation reference, the proposed method achieves less than 2 mm deviation in both seam localization and dislocation analysis, demonstrating high consistency with manual interpretation. Compared with existing automatic approaches, the proposed method provides more accurate and reliable automated dislocation analysis, significantly reducing the need for manual inspection. The proposed method enhances the automation, consistency, and reliability of shield tunnel deformation assessment and provides an effective solution for structural health monitoring. Full article

Non-contact metric inspection of metallic circular holes is essential for assembly quality control, yet remains difficult on reflective machined surfaces, where depth scale, stripe radiometry, and contour geometry degrade simultaneously. Pure monocular two-dimensional vision localizes hole boundaries efficiently but cannot resolve metric depth; multi-camera three-dimensional systems remove this ambiguity but with a heavier hardware and calibration cost; and conventional structured-light pipelines often improve stripe extraction or circle fitting in isolation, leaving the overall measurement chain fragile when reflection and edge defects co-occur. This paper proposes a monocular structured-light framework that treats sensing geometry, radiometric stripe reliability, and outlier-robust hole estimation as one coupled measurement chain. A single calibrated industrial camera is combined with an obliquely projected line laser to fuse top-view contour observation with light-plane-constrained depth recovery. Multi-exposure high-dynamic-range (HDR) fusion, adaptive Gaussian regularization, and distance-weighted gray-centroid refinement stabilize sub-pixel stripe centerlines under local saturation and uneven illumination, while geometry-aware contour screening and probabilistic multi-stage RANSAC fitting suppress burr-induced outliers during circle-parameter estimation. Experiments were conducted on a steel bolt-hole workpiece and a 6061 aluminum-alloy plate containing five holes with nominal diameters spanning approximately 30–78 mm. The original steel workpiece was evaluated for its diameter and two datum-related center-position quantities, while the five-hole plate was evaluated through three repeated optical diameter measurements and independent CMM references. For the five aluminum-alloy holes, the mean optical–CMM diameter differences ranged from 0.006 to 0.218 mm, with an average absolute difference of 0.096 mm. The results establish feasibility over the tested workpieces and calibrated measurement volume rather than generalization to arbitrary hole geometries, materials, or surface conditions. Full article

The delay-and-sum total focusing method (TFM) for ultrasonic full matrix capture (FMC) depends on accurate ray path and travel time computation. In L-shaped carbon fiber-reinforced polymer (CFRP) corner parts, elastic anisotropy, multilayer stacking, and curvature-induced ray path non-uniqueness generate strong stripe-like coherent clutter (deterministic structural echoes), degrading focusing and sizing. To address this, we search multiple physically plausible candidate ray paths and propose a phase-consistency-adaptive multi-path fusion TFM (PCA-MPF-TFM) that performs pixel-wise path selection and fusion. The method is validated using pulse-echo FMC data acquired with a water-immersion linear array from a 6.4 mm-thick L-shaped CFRP specimen containing three 3 mm-diameter polytetrafluoroethylene (PTFE) inserts; the two within the concave-side inspection region were quantitatively evaluated. Compared with conventional isotropic TFM, an edge-adjacent delamination previously masked by structural noise is consistently detected with a 9.2 dB signal-to-noise ratio (SNR) and a 0.2 mm length error. For the second delamination, the SNR improves by 25 dB and the length error decreases from 0.6 mm to 0.2 mm. Experimental results demonstrate improved defect detectability and noise robustness under curved, anisotropic, and multilayer propagation while maintaining sub-millimeter sizing accuracy. Full article

Accurate target detection in side-scan sonar imagery is important for marine resource investigation, underwater infrastructure inspection, and maritime security. However, Side-scan sonar images are often affected by low contrast, acoustic speckle, weak target boundaries, and cluttered seabed backgrounds, which make target detection particularly challenging, especially under limited training data conditions. To improve the representation of sonar-specific structures, this study proposes a multi-feature early fusion detection network, referred to as MFEF-Det, for side-scan sonar target detection. The method combines multiple handcrafted features with data-driven representations to provide complementary information. In particular, a directional contrast core response feature (DCCR) is introduced to better emphasize the echo–shadow structure commonly observed in side-scan sonar imagery. An adaptive fusion strategy is then adopted to combine multiple feature maps before feeding them into a detection network, and an attention refinement module is further employed for complex scenes to improve the discrimination between target-related regions and cluttered backgrounds. Experiments were conducted on two publicly available sonar datasets. Experiments on the KLSG and SSS-Bottom datasets demonstrate that MFEF-Det achieves 0.933 ± 0.016 mAP@0.5 and 0.866 ± 0.052 mAP@0.5, respectively. The results indicate that the proposed feature representation can improve detection performance in both relatively clean and more challenging noisy scenes. These findings suggest that incorporating sonar-specific priors can be beneficial for side-scan sonar detection in marine survey and maritime-security applications. Full article

Coffee leaf diseases impair photosynthesis, thereby degrading the chemical composition and flavor quality of coffee beans. However, these resulting quality defects are often not visible from the appearance of green beans alone, compelling e-commerce quality control to trace back to leaf disease detection at the production origin. This focus on “quality traceability” imposes requirements on computer vision technologies that fundamentally differ from those of conventional pesticide-application-oriented detection: prioritizing high precision over high recall, replacing simple classification with severity grading, and necessitating the integration of variety and origin metadata. This paper conducts a systematic review of 53 relevant studies published from January 2020 to March 2026, examining existing data resources, model architectures, and industrial adaptability through the lens of e-commerce quality traceability. Our review highlights three major findings: (1) existing datasets could be further enriched in variety labeling, severity scoring, and origin metadata; (2) current models, predominantly focused on classification, have room for closer alignment with traceability requirements regarding optimization objectives, task definitions, and output formats; and (3) cutting-edge technologies, including semantic segmentation, multimodal fusion, and visual foundation models, offer viable pathways to bridge these gaps. This review provides standardized technical evaluation criteria and clear optimization directions for origin inspection, batch grading and whole-chain traceability management of coffee agricultural e-commerce platforms. Full article

Fruit preharvest–postharvest quality assessment is essential for precision harvesting, intelligent grading, storage management, and supply-chain loss reduction. However, conventional approaches mainly rely on single-point postharvest inspection and cannot adequately capture the long-term effects of preharvest fruit phenotypes and environmental dynamics on quality formation. To address limited prediction accuracy under few-label conditions, insufficient multimodal fusion, and weak cross-orchard generalization, this study proposes FruitSSL-QNet, a self-supervised multimodal learning framework for jointly modeling preharvest fruit images, environmental sensor time series, and postharvest quality indicators. The framework employs visual masked reconstruction to learn fine-grained phenotype features, including color, texture, lenticel distribution, disease spots, and maturity patterns. Environmental temporal masked modeling is used to capture the cumulative effects of temperature, humidity, light intensity, soil moisture, and rainfall. Bidirectional cross-attention, gated fusion, and contrastive alignment are further integrated to learn complementary and semantically consistent image–environment representations. Experimental results demonstrate that FruitSSL-QNet outperforms SVM, Random Forest, XGBoost, LSTM, GRU, TCN, Transformer, and MM-Transformer across multiple quality assessment tasks. The proposed model achieves a maturity recognition accuracy of $89.6\%$, exceeding MM-Transformer by $4.4$ percentage points. Compared with the corresponding baseline results, the prediction errors for sugar content, firmness, and shelf life are reduced by $21.1\%$, $22.2\%$, and $22.5\%$, respectively. The decay-risk AUC reaches $0.921$, and the cross-site F1-score reaches $0.867$, indicating strong risk discrimination and stable generalization across orchard environments. Ablation experiments further confirm the contributions of visual self-supervision, environmental temporal self-supervision, and cross-modal alignment. Full article

Severe windthrow events can create highly favorable conditions for bark beetle outbreaks by generating large quantities of weakened breeding material. This study investigated the population response of the Mediterranean pine engraver, Orthotomicus erosus Wollaston, following a microburst windstorm that struck Marjan Forest Park, an urban Mediterranean Aleppo pine (Pinus halepensis Mill.) forest in Split, Croatia, in July 2025. Approximately 2000 m³ of damaged timber remained in the forest until February 2026, enabling assessment of colonization dynamics and outbreak development. Field inspections, laboratory analyses, and pheromone trap monitoring were conducted between July 2025 and April 2026. Insect sampling was conducted in the field through bark inspections and pheromone trap monitoring. For log inspections, bark sections were carefully removed to expose galleries and immature stages beneath the bark. Observations focused on the presence of entrance holes, boring dust, maternal galleries, larval galleries, pupal chambers, exit holes, adults, and immature developmental stages. In addition, ten pheromone traps were installed shortly after the windthrow event. Trap catches were collected regularly, and the captured insects were examined in the laboratory, where O. erosus individuals were identified and counted under a stereomicroscope. Colonization of windthrown material began within two weeks after the disturbance, and infestation rates reached 94% of examined logs by December 2025. At that point, 86% of infested logs already contained exit holes, indicating successful completion of beetle development. Pheromone traps captured a total of 131,588 adults, with more than 88% recorded during spring 2026 and a pronounced maximum recorded catch occurring on 8 April 2026. The appearance of newly attacked standing trees in spring (April) 2026 confirmed rapid population expansion from fallen material to living hosts. Results demonstrate the exceptional outbreak potential of O. erosus following extreme climatic disturbances and emphasize the importance of rapid sanitation measures and continuous monitoring in Mediterranean pine forests under climate change conditions. Delayed salvage logging after severe windthrow can rapidly trigger an outbreak of O. erosus. Full article

Termite infestation poses a critical threat to ancient timber structures because hidden in-wood activity can cause progressive structural decay before visible surface symptoms appear. Through-wood termite detection remains challenging because termite-induced electromagnetic responses are weak, small-scale, and vulnerable to timber attenuation and multipath propagation. To address this problem, this study presents one of the first investigations to formulate through-wood termite detection as a multi-frequency wireless sensing and image–wireless fusion problem for non-destructive heritage timber inspection. We propose a Reliability-Aware Image–Wireless Fusion Network (RA-IWFNet), in which the image branch captures high-resolution surface-level visual cues while the dual-band wireless branch integrates complementary mmWave radar micro-motion responses and Wi-Fi Channel State Information (CSI) channel variations. A learnable temperature-scaled fusion gate estimates input-dependent image and wireless contributions and constructs a normalized fused representation for four-class recognition, including Termite, Lyctidae, Human, and None. Here, reliability is operationally defined as learned input-adaptive relative modality contribution rather than explicit uncertainty or signal-quality estimation. RA-IWFNet is evaluated under two complementary protocols: a field-motivated protocol with joint visual and wireless degradation and a synchronized verification protocol using physically co-acquired multimodal samples. Across repeated training runs, RA-IWFNet achieves $81.91\pm 1.33\%$ accuracy and $81.96\pm 1.27\%$ Macro-F1 under field-mixed visual degradation and moderate wireless degradation. On the synchronized verification subset, gated fusion achieves $91.53\pm 2.44\%$ accuracy and $91.62\pm 2.44\%$ Macro-F1, yielding higher mean performance than single-modality and non-adaptive fusion baselines. Feature-space, error-correction, and gate-temperature analyses further support the effectiveness of adaptive modality integration. These results provide controlled laboratory feasibility evidence and suggest that multi-frequency wireless sensing combined with adaptive image–wireless fusion offers a promising non-invasive pathway toward practical through-wood termite inspection in heritage timber structures. Full article

Solder-joint detection and localization on large, complex printed circuit boards (PCBs) remain challenging because PCB images often contain unevenly distributed features, nonuniform illumination, specular reflection, scale variation and geometric distortion. Conventional SIFT-based registration methods usually use fixed matching parameters and therefore cannot adapt well to regions with different component densities. To address this limitation, this study proposes a hardware–software integrated PCB image registration framework that combines a robotic end effector with a locally adaptive K-nearest neighbor (LAKNN) strategy and SIFT descriptors. The hardware platform provides stable image acquisition through a lifting mechanism and ring-light illumination, while the software module adjusts the neighbor-search space according to local feature density and matching confidence. Experimental results show that the proposed LAKNN method achieves 90.8% inlier-match accuracy in the ablation experiment and improves registration robustness under height, region and viewpoint variations. The proposed framework provides a practical basis for automated PCB inspection and robotic soldering alignment. Full article

Existing bridges represent a significant challenge for structural engineering due to aging, increased traffic demands, and lack of original design information. This paper presents an investigation of the structural behavior of the Miguel Hidalgo Bridge, located in Culiacán, Sinaloa, México, which has remained in service for over a century. A preliminary visual inspection identified deterioration mechanisms, including cracking, exposed reinforcement, damaged bearings, and localized structural deformations. Given the absence of original design documentation, a structural health monitoring (SHM) strategy was implemented to assess its condition. The SHM methodology integrates field instrumentation using accelerometers for operational modal analysis under ambient vibration and GPS receivers for displacement measurements, complemented by a geometric survey and material characterization through non-destructive techniques. These data were used to develop and calibrate a three-dimensional finite element (FE) model, which was employed to evaluate structural performance under service loads in accordance with AASHTO provisions. Results indicate that the bridge satisfies serviceability criteria; however, certain components exhibit deficiencies in load-carrying capacity. These findings highlight the importance of integrating monitoring data with numerical modeling to support decision-making related to maintenance, rehabilitation, and extension of service life in existing bridge infrastructure. Full article

For final-state quality inspection of three-dimensional stamped hairpin windings, existing studies still lack multimodal methods that integrate mechanical geometric constraints, prior-guided fusion, and decision interpretability. This study proposes a span-prior-guided explainable multimodal neural network method and develops SPIMA-Net. The final-state images were acquired at a fixed inspection station with a fixed camera position and imaging angle under a CCD vision light source. Final-state images are used as visual inputs, while geometric priors are constructed from span measurements and model-type information. A visual branch and a span branch extract image and prior features, and a span-prior-assisted gating mechanism modulates visual features to enable collaborative fusion. Experimental results show that SPIMA-Net achieves an accuracy of 98.14%, an F1-score of 96.55%, and an AUC of 0.9983 on the test set. Its nonconforming-class F1-score is improved by 10.44, 3.04, 6.27, 1.38, and 0.66 percentage points over the image-only, span-only, direct-fusion, SE-fusion, and CBAM-fusion models, respectively, while the total number of misclassifications decreases to five. Interpretability analysis shows that the model mainly focuses on span openings, end profiles, and local abnormal regions. Relative and absolute span deviations are identified as the main mechanical geometric factors affecting final-state quality classification. The proposed method provides a neuro-mechanical fusion approach that demonstrates high discriminative performance and engineering interpretability for hairpin winding quality inspection on the investigated industrial dataset. Full article

Shallow defects in ferromagnetic components may produce weak and unstable magnetic flux leakage responses in compact inspection devices, whereas an inductive response alone does not provide the same depth-related magnetic information. To obtain complementary defect information, this study proposes a magnetic–inductive dual-channel inspection method that integrates an MLX90393 three-axis digital magnetic sensor with a PCB planar spiral coil and an LDC1612 inductance-to-digital converter. Because the two sensing units are physically separated on the detection board, peak-position offset analysis and spatial registration were introduced to associate their responses to the same defect region. Controlled experiments were conducted on a laboratory pipeline inspection platform using a Q235 defect specimen installed at the internal inspection position of the pipe. The defects were machined on the inner surface, which was also the inspection surface. For each defect, five motor-driven axial scans were performed at 10 mm/s, with 400 samples acquired at 100 Hz during each 4 s scan. Response amplitude, signal-to-noise ratio, peak position, and repeatability were evaluated. For the square-hole defects, the magnetic response amplitude decreased from 675.0 to 98.5 a.u. as the depth ratio decreased from 50% to 10%, while the magnetic-channel SNR decreased from 30.5 to 4.7. For the 10% depth defect, the inductive channel retained an average SNR of 434.4 and a coefficient of variation of 0.23%, providing a stable auxiliary response. However, the pre-registration peak-position offset measured for this shallow defect was $21.2\pm 20.9$ sampling points, indicating substantial uncertainty in using a single magnetic extremum for scan-specific registration when the magnetic response was weak. These results demonstrate that the two channels provide different and complementary information under controlled laboratory conditions, while further validation is required for irregular corrosion, other ferromagnetic components, and practical inspection conditions. Full article

Background/Objectives: Deciding between active surveillance and treatment for prostate cancer patients often requires accurate risk assessment of prostate tumors detected with multi-parametric MRI. Conventionally, radiologists visually inspect MRI and use scoring procedures such as PI-RADS to help assess the scans. More recently, artificial intelligence (AI) applied to MRI has allowed for supplementation and is complementary to clinical assessment. However, AI is computationally expensive and severely saps scarce energy and water resources and requires special processing components, requiring alternate approaches that require less computation and fewer resources. The novel, simpler spectral/statistics approach that mimics color vision was previously successfully applied in a number of retrospective pilot studies of bi-parametric MRI of prostate cancer. The novel approach needs far fewer resources, is less computationally intensive, and is simpler than artificial intelligence to evaluate prostate tumors. However, these earlier spectral/statistics pilot studies required intervention by an analyst and too much time for implementation in future large patient studies that are needed to validate the novel approach. This retrospective pilot study further developed, applied, and tested new automation tools to expedite simpler spectral statistical techniques that need fewer resources to evaluate prostate tumors on multi-parametric MRI. Methods: Automated spatial registration, automated prostate organ segmentation, automated blob generation and selection for spectral signatures derived from the apparent diffusion coefficient, high-B-value DWI, and T2 MRI were performed on 76 consecutive patients in the PI-CAI cohort in this retrospective pilot study. The signal-to-clutter ratio (SCR) was computed using target signatures and the processed statistical metrics of the registered prostate bi-parametric MRI. The processed SCR, spectral/spatial features of blobs and clinical metrics predict clinically significant prostate cancer using multivariate logistic regression. The proposed method was assessed using the area under the curve (AUC) from the receiver operating characteristic curve. Results: AUC values of >0.90 were achieved by combining the SCR with blob and clinical metrics. Increasing the number of non-congruent, independent variables resulted in higher AUC scores. Restricting analysis to blob volumes > 0.1 cm³ achieved higher AUC values. The additional total savings in time by applying the new automation tools reduced the processing time by 80 to 170 min for 10 patients. Implementing the new automation tools resulted in an overall processing time of 40 to 80 min per 10 patients. Conclusions: Automating the spectral/statistics approach resulted in AUCs not inferior to those obtained from AI. The automation achieved sufficiently high AUCs and also reduced processing times, warranting future assessments in large patient cohorts. Full article

Navigation mark inspections are critical to ensuring maritime navigation safety, and the efficiency of unmanned aerial vehicle (UAV) inspection path planning directly affects inspection costs and operations. This problem is formulated as a Traveling Salesman Problem (TSP), and the traditional Grey Wolf Optimizer (GWO) has limitations in discrete optimization, including weak search capabilities, simple neighborhood structures, and poor local optimization. To address these issues, this paper proposes an improved Grey Wolf Optimizer (IGWO). First, this paper introduces three neighborhood search operators: reverse, insertion, and swap. Second, an adaptive step size mechanism based on Euclidean distance is designed. Third, the 3-opt local optimization algorithm is integrated. Finally, experiments are conducted using real navigation mark data from Pingtan and Tianjin, and IGWO is compared with traditional algorithms. Results show that IGWO effectively adapts GWO to discrete spaces and achieves optimal paths across datasets of varying scales. Its path length reduction rates improve by 0.51% to 58.01% over the other seven algorithms. These findings provide efficient UAV path planning solutions for navigation mark inspection and offer technical support for smart maritime supervision systems. Full article

Cracks indicate the deterioration of civil engineering structures, and early detection through regular inspection is crucial for structural safety. However, traditional manual inspection is time- and labor-intensive and depends heavily on the inspector’s expertise. Although deep learning-based crack detection has been actively studied, most prior studies focus on a single environment such as asphalt or concrete, and research integrating asphalt, concrete, and tunnel environments remains scarce. In this study, we propose a Dense U-Net++-based model trained on a unified dataset of drone-captured crack images across all three environments. The model combines dense connections with U-Net++’s nested skip pathways to mitigate the semantic gap in the encoder–decoder structure, enabling effective feature fusion and precise restoration of fine crack boundaries. Focal loss addresses the severe class imbalance between background and crack pixels, and area-based postprocessing suppresses spurious detections. The proposed method achieved a precision of 95.47%, a recall of 92.35%, an F1-score of 93.91%, and an IoU of 88.58%, outperforming both the baseline Dense U-Net++ and Mask R-CNN. Qualitative evaluation confirmed reliable detection across rough asphalt textures, striped concrete patterns, and low illumination in tunnels. These results indicate that the proposed framework achieves promising crack segmentation performance for automated structural inspection. Full article