Published: Sept. 13, 2024
Language: Английский
Journal of Real-Time Image Processing, Journal Year: 2024, Volume and Issue: 21(2)
Published: Jan. 30, 2024
Language: Английский
Citations
6
Applied Sciences, Journal Year: 2024, Volume and Issue: 14(11), P. 4766 - 4766
Published: May 31, 2024
Timely and accurately detecting personal protective equipment (PPE) usage among workers is essential for substation safety management. However, traditional algorithms encounter difficulties in substations due to issues such as varying target scales, intricate backgrounds, many model parameters. Therefore, this paper proposes MEAG-YOLO, an enhanced PPE detection built upon YOLOv8n. First, the incorporates Multi-Scale Channel Attention (MSCA) module improve feature extraction. Second, it newly designs EC2f structure with one-dimensional convolution enhance fusion efficiency. Additionally, study optimizes Path Aggregation Network (PANet) learning of multi-scale targets. Finally, GhostConv integrated optimize operations reduce computational complexity. The experimental results show that MEAG-YOLO achieves a 2.4% increase precision compared YOLOv8n, 7.3% reduction FLOPs. These findings suggest effective identifying complex scenarios, contributing development smart grid systems.
Language: Английский
Citations
5
Applied Sciences, Journal Year: 2024, Volume and Issue: 14(17), P. 7923 - 7923
Published: Sept. 5, 2024
Wearing safety helmets at construction sites is a major measure to prevent accidents, so it essential supervise and ensure that workers wear helmets. This requires high degree of real-time performance. We improved the network structure based on YOLOv7. To enhance performance, we introduced GhostModule after comparing various modules create new efficient generates more feature mappings with fewer linear operations. SE blocks were several attention mechanisms highlight important information in image. The EIOU loss function was speed up convergence model. Eventually, constructed model EGS-YOLO. EGS-YOLO achieves mAP 91.1%, 0.2% higher than YOLOv7, inference time 13.3% faster YOLOv7 3.9 ms (RTX 3090). parameters computational complexity are reduced by 37.3% 33.8%, respectively. enhanced performance while maintaining original precision can meet actual detection requirements.
Language: Английский
Citations
4
Electronics, Journal Year: 2024, Volume and Issue: 13(23), P. 4589 - 4589
Published: Nov. 21, 2024
A safety helmet is indispensable personal protective equipment in high-risk working environments. Factors such as dense personnel, varying lighting conditions, occlusions, and different head postures can reduce the precision of traditional methods for detecting helmets. This paper proposes an improved YOLOv8n detection model, YOLOv8-ADSC, to enhance performance complex In this firstly, Adaptive Spatial Feature Fusion (ASFF) Deformable Convolutional Network version 2 (DCNv2) are used head, enabling network more effectively capture multi-scale information target; secondly, a new layer small targets incorporated sensitivity smaller targets; finally, Upsample module replaced with lightweight up-sampling Content-Aware ReAssembly Features (CARAFE), which increases perception range, reduces loss caused by up-sampling, improves robustness target detection. The experimental results on public Safety-Helmet-Wearing-Dataset (SHWD) demonstrate that, comparison original [email protected] YOLOv8-ADSC has increased 2% all classes, reaching 94.2%, [email protected]:0.95 2.3%, 62.4%. be better suited
Language: Английский
Citations
4
Measurement, Journal Year: 2025, Volume and Issue: unknown, P. 117572 - 117572
Published: April 1, 2025
Language: Английский
Citations
0
Journal of Real-Time Image Processing, Journal Year: 2024, Volume and Issue: 21(4)
Published: July 5, 2024
Language: Английский
Citations
2
IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 126313 - 126328
Published: Jan. 1, 2024
Wearing a safety helmet is an essential aspect of production management. To address the issue low detection accuracy existing algorithms for small targets and complex scenes, we propose YOLOv8n-ASF-DH model (DH stands Dynamic Head), which based on YOLOv8n integrates Attentional Scale Sequence Fusion (ASF) structure Head (DyHead). In backbone layer, Triplet Attention mechanism incorporated to enhance model's focus features targets. neck ASF efficiently fuses different level output extracted by network, enhancing overall feature representation capability model. head, DyHead adjusts relationships between layers, performance scales scenes. Additionally, adopting Focal-EIoU loss function balances contributions high-quality low-quality samples in calculation. Experimental results demonstrate that improved enhances scenes target scenarios. Compared model, achieved improvement 3.066% 3.883% recall. exhibited increase 2.584% mAP 0.5 6.131% 0.5:0.95. with other mainstream object algorithms, significantly tasks.
Language: Английский
Citations
2
Published: Feb. 19, 2024
The COVID-19 coronavirus pandemic has generated a global health crisis in all Worldwide. According to the World Health Organization (WHO), protection against infection is an essential countermeasure. one of most effective countermeasures wearing facial mask which imperative our everyday activities, particularly communal settings, mitigate transmission illness. In this study, we have enhanced architectural design YOLOv5n v6.0 for face detection by constructing modified model known as Proposal model. primary objective modification enhance feature extraction and prediction capabilities proposal, outline integration residual network (ResNet) backbone into architecture replacing first three layer v6.0with ResNet_Stem module ResNet_Block replace Spatial Pyramid Pooling Fast (SPPF) original with Pooling-Cross Stage Partial (SPPCSP) modules combines SPP CSP create that both efficient. proposal carefully curated set anchor configurations tailored specific requirements small object detection. MJFR dataset was used testing evaluation proposed consist 23,621 images collected authors paper. performance models evaluated using following metrics: mean average precision (mAP50), mAP50-95, recall (R) (P). We conclude outperforms terms accuracy mAP50, mAP50-95 metric measure
Language: Английский
Citations
1
Journal of Real-Time Image Processing, Journal Year: 2024, Volume and Issue: 21(4)
Published: June 25, 2024
Language: Английский
Citations
1
Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown
Published: Jan. 11, 2024
Abstract Wearing inspection safety equipment such as insulating gloves and helmets is an important guarantee for safe power operations. Given the low accuracy of traditional helmet-wearing detection algorithm problems missed false detection, this paper proposes improved wearing model named RepGFPN-YOLOv5 based on YOLOv5. This first uses K-Means + to analyze data set Anchor parameter size re-clustering optimize target anchor box size; secondly, it neck network (Efficient Reparameterized Generalized Feature Pyramid Network, RepGFPN), which combines efficient layer aggregation ELAN re-parameterization mechanism), reconstruct YOLOv5 improve feature fusion ability network; reintroduce coordinate attention mechanism (Coordinate Attention, CA) focus small information; finally, use WIoU_Loss loss function reduce prediction errors. Experimental results show that achieves increase 2.1% mAP value 2.3% compared with original network, speed reaches 89FPS.The code: https://github.com/CVChenXC/RepGFPN-YOLOv5.git.
Language: Английский
Citations
0