An Optimized YOLOv11 Framework for the Efficient Multi-Category Defect Detection of Concrete Surface

Sensors, Journal Year: 2025, Volume and Issue: 25(5), P. 1291 - 1291

Published: Feb. 20, 2025

Thoroughly and accurately identifying various defects on concrete surfaces is crucial to ensure structural safety prolong service life. However, in actual engineering inspections, the varying shapes complexities of challenge insufficient robustness generalization mainstream models, often leading misdetections under-detections, which ultimately jeopardize safety. To overcome disadvantages above, an efficient defect detection model called YOLOv11-EMC (efficient multi-category detection) proposed. Firstly, ordinary convolution substituted with a modified deformable efficiently extract irregular features, model’s are significantly enhanced. Then, C3k2module integrated revised dynamic module, reduces unnecessary computations while enhancing flexibility feature representation. Experiments show that, compared Yolov11, Yolov11-EMC has improved precision, recall, mAP50, F1 by 8.3%, 2.1%, 4.3%, 3% respectively. Results drone field tests that successfully lowers false under-detections simultaneously increasing accuracy, providing superior methodology tasks require tangible flaws practical applications.

Language: Английский

---

Research on UAV Aerial Imagery Detection Algorithm for Mining-Induced Surface Cracks Based on Improved YOLOv10

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: May 19, 2025

UAV-based aerial imagery plays a vital role in detecting surface cracks mining-induced areas for geological disaster early warning and safe production. However, detection is challenged by small crack size, complex morphology, large scale variation, uneven spatial distribution, further exacerbated UAVs' limited onboard computational capacity. To tackle these issues, we introduce an efficient lightweight small-target model, namely YOLO-LSN, which built upon the optimized YOLO architecture.Firstly, Lightweight Dynamic Alignment Detection Head (LDADH) multi-scale feature fusion, precise alignment, dynamic receptive field adjustment, optimizing extraction. Secondly, Small Object Feature Enhancement Pyramid (SOFEP) enhances detail representation of backgrounds.Furthermore, propose weighted combination strategy Normalized Wasserstein Distance (NWD) IoU loss, balancing sensitivity to zero-overlap instances robustness against deviations, thereby improving localization accuracy generalization capability. Experiments show 12% [email protected] improvement 17% reduction parameters on self-built mining dataset, with validation VisDrone2019 ([email protected]: 0.422, + 11.6%), Validating its effectiveness small-object detection, model offers efficient, reliable solution hazard safety.

Language: Английский

---

Vision-Based Localization Method for Picking Points in Tea-Harvesting Robots

Sensors, Journal Year: 2024, Volume and Issue: 24(21), P. 6777 - 6777

Published: Oct. 22, 2024

To address the issue of accurately recognizing and locating picking points for tea-picking robots in unstructured environments, a visual positioning method based on RGB-D information fusion is proposed. First, an improved T-YOLOv8n model proposed, which improves detection segmentation performance across multi-scale scenes through network architecture loss function optimizations. In far-view test set, accuracy tea buds reached 80.8%; near-view mAP0.5 values stem bounding boxes masks 93.6% 93.7%, respectively, showing improvements 9.1% 14.1% over baseline model. Secondly, layered servoing strategy near far views was designed, integrating RealSense depth sensor with robotic arm cooperation. This identifies region interest (ROI) bud view fuses mask data to calculate three-dimensional coordinates point. The experiments show that this achieved point localization success rate 86.4%, mean measurement error 1.43 mm. proposed recognition reduces fluctuations, providing technical support intelligent rapid premium tea.

Language: Английский

---

Mixed Reality-Based Concrete Crack Detection and Skeleton Extraction Using Deep Learning and Image Processing

Electronics, Journal Year: 2024, Volume and Issue: 13(22), P. 4426 - 4426

Published: Nov. 12, 2024

Advancements in image processing and deep learning offer considerable opportunities for automated defect assessment civil structures. However, these systems cannot work interactively with human inspectors. Mixed reality (MR) can be adopted to address this by involving inspectors various stages of the process. This paper integrates You Only Look Once (YOLO) v5n YOLO v5m Canny algorithm real-time concrete crack detection skeleton extraction a Microsoft HoloLens 2 MR device. The demonstrates superior mean average precision (mAP) 0.5 speed, while achieves highest mAP 0.95 among other v5 also outperforms Sobel Prewitt edge detectors F1 score. developed MR-based system could not only employed but utilized automatic recording location specifications cracks further analysis future re-inspections.

Language: Английский

---