2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP), Год журнала: 2024, Номер unknown, С. 167 - 173
Опубликована: Апрель 19, 2024
Язык: Английский
2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP), Год журнала: 2024, Номер unknown, С. 167 - 173
Опубликована: Апрель 19, 2024
Язык: Английский
Drones, Год журнала: 2024, Номер 8(3), С. 84 - 84
Опубликована: Фев. 28, 2024
In smart cities, effective traffic congestion management hinges on adept pedestrian and vehicle detection. Unmanned Aerial Vehicles (UAVs) offer a solution with mobility, cost-effectiveness, wide field of view, yet, optimizing recognition models is crucial to surmounting challenges posed by small occluded objects. To address these issues, we utilize the YOLOv8s model Swin Transformer block introduce PVswin-YOLOv8s for detection based UAVs. Firstly, backbone network incorporates global feature extraction object Secondly, challenge missed detections, opt integrate CBAM into neck YOLOv8. Both channel spatial attention modules are used in this addition because how well they extract information flow across network. Finally, employ Soft-NMS improve accuracy occlusion situations. increases performance manages overlapped boundary boxes well. The proposed reduced fraction objects overlooked enhanced performance. Performance comparisons different YOLO versions ( example YOLOv3 extremely small, YOLOv5, YOLOv6, YOLOv7), YOLOv8 variants (YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l), classical detectors (Faster-RCNN, Cascade R-CNN, RetinaNet, CenterNet) were validate superiority model. efficiency was confirmed experimental findings, which showed 4.8% increase average (mAP) compared VisDrone2019 dataset.
Язык: Английский
Процитировано
43IEEE Geoscience and Remote Sensing Letters, Год журнала: 2025, Номер 22, С. 1 - 5
Опубликована: Янв. 1, 2025
Язык: Английский
Процитировано
3Sensors, Год журнала: 2024, Номер 24(9), С. 2905 - 2905
Опубликована: Май 1, 2024
Underwater visual detection technology is crucial for marine exploration and monitoring. Given the growing demand accurate underwater target recognition, this study introduces an innovative architecture, YOLOv8-MU, which significantly enhances accuracy. This model incorporates large kernel block (LarK block) from UniRepLKNet to optimize backbone network, achieving a broader receptive field without increasing model’s depth. Additionally, integration of C2fSTR, combines Swin transformer with C2f module, SPPFCSPC_EMA blends Cross-Stage Partial Fast Spatial Pyramid Pooling (SPPFCSPC) attention mechanisms, notably improves accuracy robustness various biological targets. A fusion DAMO-YOLO further multi-scale feature extraction capabilities in neck. Moreover, adoption MPDIoU loss function, designed around vertex distance, effectively addresses challenges localization boundary clarity organism detection. The experimental results on URPC2019 dataset indicate that YOLOv8-MU achieves [email protected] 78.4%, showing improvement 4.0% over original YOLOv8 model. URPC2020 dataset, it 80.9%, and, Aquarium reaches 75.5%, surpassing other models, including YOLOv5 YOLOv8n, thus confirming wide applicability generalization our proposed improved architecture. Furthermore, evaluation demonstrates leading performance (SOTA), 88.1%, verifying its superiority dataset. These highlight broad across datasets.
Язык: Английский
Процитировано
6Computers and Electronics in Agriculture, Год журнала: 2024, Номер 225, С. 109344 - 109344
Опубликована: Авг. 22, 2024
Язык: Английский
Процитировано
6Опубликована: Апрель 11, 2024
Underwater visual detection technology plays a pivotal role in fields such as marine exploration. With the increasing demand for underwater monitoring, quest efficient and reliable methods target recognition has become particularly significant. To address this requirement, study developed an innovative object architecture based on YOLOv8, named YOLOv8-MU, aimed at significantly enhancing accuracy.By integrating LarK module proposed UniRepLKNet to optimize backbone network, YOLOv8-MU aims achieve larger receptive field without model’s depth. Further, research introduces C2fSTR, method that combines Swin Transformer with C2f module. Additionally, we have incorporated SPPFCSPC_EMA module, which Cross-Stage Partial Fast Spatial Pyramid Pooling (SPPFCSPC) attention mechanisms, improving accuracy robustness of various biological targets. Moreover, by introducing fusion block DAMO-YOLO into neck model, further enhanced capability multi-scale feature extraction. Finally, adoption MPDIoU loss function, designed around vertex distance, effectively tackles challenges localization boundary clarity organism detection. Experimental results URPC2019 dataset demonstrate model achieved [email protected] 78.4%, marking improvements 5.6%, 1.1%, 4.0% over YOLOv5s, YOLOv7, YOLOv8n respectively, indicating leading performance (SOTA) method. On other hand, evaluation URPC2020 confirmed generalization architecture, its reaching 80.4%, surpassing models including YOLOv5x YOLOv8n, showcasing wide applicability our improved architecture.
Язык: Английский
Процитировано
5Measurement Science and Technology, Год журнала: 2024, Номер 35(9), С. 095402 - 095402
Опубликована: Апрель 23, 2024
Abstract Object detection in remote sensing imagery exhibits difficulties due to complex backgrounds, diverse object scales, and intricate spatial context relationships. Motivated by the problems mentioned above, this paper introduces AeroDetectNet, a novel lightweight high-precision network custom-designed for aerial scenarios, building upon YOLOv7-tiny algorithm. It enhances performance through four key improvements: normalized Wasserstein distance consistent size sensitivity, Involution module reduced background noise, self-designed RCS-biformer better interpretation, WF-CoT SPPCSP feature pyramid improved map weighting capture. Ablation studies conducted on hybrid dataset composed of three open-source datasets (including NWPU VHR-10 images, RSOD VisDrone UAV images) have demonstrated effectiveness improvements specifically small-size detection. Visualizations Grad-CAM further demonstrate AeroDetectNet’s capacity extract focus features. Upon individual testing across datasets, AeroDetectNet has successfully its ability identify objects images with smaller pixel area. Through experimental comparisons other related studies, achieved competitive mAP while maintaining fewer model parameters, highlighting highly accurate properties.
Язык: Английский
Процитировано
5Scientific Reports, Год журнала: 2024, Номер 14(1)
Опубликована: Авг. 13, 2024
Abstract Defect detection in pharmaceutical blister packages is the most challenging task to get an accurate result detecting defects that arise tablets while manufacturing. Conventional defect methods include human intervention check quality of within packages, which inefficient, time-consuming, and increases labor costs. To mitigate this issue, YOLO family primarily used many industries for real-time continuous production. enhance feature extraction capability reduce computational overhead a environment, CBS-YOLOv8 proposed by enhancing YOLOv8 model. In CBS-YOLOv8, coordinate attention introduced improve capturing spatial cross-channel information also maintaining long-range dependencies. The BiFPN (weighted bi-directional pyramid network) fusion at each convolution layer avoid more precise loss. model's efficiency enhanced through implementation SimSPPF (simple pooling fast), reduces demands model complexity, resulting improved speed. A custom dataset containing defective tablet images train performance then evaluated comparing it with various other models. Experimental results on reveal achieves mAP 97.4% inference speed 79.25 FPS, outperforming SESOVERA-ST saline bottle fill level monitoring achieved mAP50 99.3%. This demonstrates provides optimized inspection process, enabling prompt correction defects, thus bolstering assurance practices manufacturing settings.
Язык: Английский
Процитировано
5Опубликована: Янв. 1, 2024
This study proposes a UAV-based remote measurement method for accurately locating pedestrians and other small targets within reservoir dams. To address imprecise coordinate information in areas after prolonged operations, transformation converting UAV coordinates into the local system without relying on preset parameters is introduced, accomplished by integrating Structure from Motion (SfM) algorithm to calculate parameters. An improved YOLOv8 network introduced high-precision detection of pedestrian targets, complemented laser rangefinder facilitate accurate 3D varying postures positions. Furthermore, integration thermal infrared camera facilitates localization potential seepage. Experimental validation application across two real dams confirm accuracy applicability proposed approach, demonstrating efficiency routine surveillance strategy proving its establish electronic fences enhance maintenance operations.
Язык: Английский
Процитировано
3The Journal of Supercomputing, Год журнала: 2025, Номер 81(2)
Опубликована: Янв. 21, 2025
Язык: Английский
Процитировано
0The Journal of Supercomputing, Год журнала: 2025, Номер 81(5)
Опубликована: Апрель 16, 2025
Язык: Английский
Процитировано
0