Pith. sign in

REVIEW 3 major objections 6 minor 45 references

Foreign-Object Detection in High-Voltage Transmission Line Based on Improved YOLOv8m

T0 review · 3 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read An improved YOLOv8m detector reaches 95.5% mAP_0.5 and 91.9% recall for foreign objects on high-voltage transmission lines, up 2.7 and 6 percentage points over the base model.

desk verdict Routine YOLOv8m combo paper with a real six-class dataset, but the ablation table contradicts the headline mAP_0.5:0.95 gain. read the letter →

arxiv 2502.07175 v1 pith:HHF43BVO submitted 2025-02-11 cs.CV cs.AI

Zhenyue Wang , Guowu Yuan , Hao Zhou , Yi Ma , Yutang Ma This is my paper · ORCID
classification cs.CVcs.AI
keywords transmissionlinesforeign-objectdetectionYOLOGAMSPPCSPCFocal-EIoUdataaugmentation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Foreign objects such as kites, bird nests, and plastic trash on high-voltage transmission lines can cause short circuits, and manual inspection is slow. This paper claims that an improved YOLOv8m detector, trained on a proprietary six-class dataset of aerial transmission-line images, finds these objects more reliably than the base model and than seven other detectors. Three targeted changes carry the claim: a Global Attention Module in the backbone to handle occluded objects, an SPPCSPC module replacing SPPF for multi-scale features, and a Focal-EIoU loss to rebalance high- and low-quality samples. The reported result is that mAP_0.5 rises from 92.8% to 95.5%, mAP_0.5:0.95 from 76.4% to 80.4%, and recall from 85.9% to 91.9% relative to YOLOv8m. If accurate, the improvement means fewer missed foreign objects in automated patrol images and less follow-up work for human inspectors.

What carries the argument

Three named components carry the argument. GAM (Global Attention Mechanism) is a channel-spatial attention module placed in the backbone to emphasize occluded target features. SPPCSPC is a spatial pyramid pooling module with two parallel streams, one using multi-scale max-pooling and one using convolutional processing, that replaces YOLOv8m's SPPF block to widen the receptive field for multi-scale objects. Focal-EIoU is a bounding-box regression loss that down-weights easy low-quality samples and up-weights harder high-quality samples, with a focal parameter $\gamma$ controlling suppression. The ablation study shows that each component shifts the metrics, and the full combination yields the reported mAP_0.5 of 95.5%, mAP_0.5:0.95 of 80.4%, and recall of 91.9%.

What would settle it

Re-running the full ablation with several random seeds and reporting means and standard deviations would settle the central claim: if the full model's mAP_0.5:0.95 does not reliably beat the values reached by GAM plus Focal-EIoU alone, the paper's attribution of the 4-point gain to the full combination would not hold.

Watch

Extended reading notes

Core claim

The paper claims that combining three modifications into the YOLOv8m detector produces a foreign-object detection model for high-voltage transmission lines that beats both its base model and all compared mainstream detectors. GAM inserted into the backbone focuses the network on occluded foreign objects; replacing SPPF with SPPCSPC strengthens multi-scale feature extraction under scale variation and complex backgrounds; and Focal-EIoU loss weights high-quality bounding-box regression samples more heavily. On the paper's six-class dataset (trash, twig, nest, kite, bird, balloon) with data augmentation, the complete model reaches mAP_0.5 of 95.5%, mAP_0.5:0.95 of 80.4%, and recall of 91.9% in single-run experiments, which the paper reports as 2.7, 4, and 6 percentage-point gains over YOLOv8m; here mAP_0.5 is mean average precision at IoU threshold 0.5, and mAP_0.5:0.95 averages IoU thresholds from 0.5 to 0.95. The paper further claims that these are the highest mAP_0.5, mAP_0.5:0.95, and recall values among Faster R-CNN, Sparse R-CNN, Cascade R-CNN, YOLOv5m, YOLOv6m, YOLOv7, and YOLOv8m, while staying second-fastest in inference speed.

Load-bearing premise

The reported gains rest on the assumption that the single-run evaluation numbers in Tables 2 and 3 are accurate and representative, rather than reflecting run-to-run variance or a transcription error.

Editorial extensions

If this is right

  • If the numbers hold, the improved detector can be used in periodic UAV-based patrols to flag kites, balloons, nests, twigs, birds, and trash on transmission lines with fewer misses than the base model.
  • The higher recall (91.9% versus 85.9%) directly reduces the miss-detection rate, and in the paper's inspection workflow, where detections are still checked manually, that cuts the workload of inspection personnel.
  • The comparative results indicate that on this dataset the one-stage YOLO family outperforms the two-stage R-CNN family, with the improved model holding the top mAP_0.5, mAP_0.5:0.95, and recall.
  • Data augmentation plus the improved model gives the best per-class results, most notably lifting the difficult 'twig' category from an mAP_0.5 of 76.1% to 90.4% before augmentation and to 99.5% with augmentation.
  • The paper states that detection speed is not yet sufficient for direct deployment on terminal CPU devices, so a lighter variant such as YOLOv8n is the planned next step.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An extension of this work would be a seed-averaged rerun of the ablation table; the non-monotonic values across rows leave open whether the 4-point mAP_0.5:0.95 gain attributed to the full combination is stable or mainly run-to-run variance.
  • A testable extension is to apply the same three modifications to a different transmission-line foreign-object dataset, since the current dataset is proprietary and from a single region; positive replication would support portability, and failure to replicate would point to dataset-specific effects.
  • The authors' stated plan to move to YOLOv8n suggests benchmarking the same GAM-SPPCSPC-Focal-EIoU recipe on the smaller model, measuring how much accuracy is traded for the CPU deployability they say is needed.
  • The attention-visualization comparisons imply that GAM is the component responsible for improving occluded-object focus; an ablation that scores only heavily occluded test images would test that attribution directly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes an improved YOLOv8m model for detecting six types of foreign objects (trash, twigs, nests, kites, birds, balloons) on high-voltage transmission lines. The three modifications are: inserting a Global Attention Module (GAM) into the backbone to address occlusion, replacing SPPF with SPPCSPC to improve multi-scale feature extraction, and using the Focal-EIoU loss to handle high- and low-quality sample imbalance. Experiments are conducted on a private dataset from Yunnan Power Grid, and the paper reports that the full model raises mAP_0.5 from 92.8% to 95.5%, mAP_0.5:0.95 from 76.4% to 80.4%, and recall from 85.9% to 91.9% relative to YOLOv8m. The model is also compared against Faster RCNN, Sparse RCNN, Cascade RCNN, YOLOv5m, YOLOv6m, YOLOv7, and YOLOv8m.

Significance. The problem is practically important for automated power-line inspection, and the paper evaluates on a real industrial dataset rather than a synthetic benchmark. The comparison set includes several recent detectors, and the attention visualizations give qualitative support for the GAM component. The core limitation is statistical: all reported results are single-run metrics with no variance, seeds, or significance tests, and the ablation in Table 2 is non-monotonic in a way that directly undercuts the attribution of the headline gains to the full combination. If the numbers are accurate and reproducible, the work is a modest but useful engineering improvement; as presented, the central claim is not established. Reproducibility is further limited by the private dataset and the absence of code, although the authors state an intention to release the dataset after approval.

major comments (3)
  1. [5.3.3, Table 2] The ablation results do not support the paper's attribution of the headline gains to the full combination. For mAP_0.5:0.95, the reported values are 76.4 for YOLOv8m, 78.5 for GAM, 73.71 for SPPCSPC, 75.12 for Focal-EIoU, 78.3 for GAM+SPPCSPC, 82.1 for SPPCSPC+Focal, 84.5 for GAM+Focal, and 80.4 for the full model. The full model is 4.1 points below GAM+Focal and 1.7 points below SPPCSPC+Focal on this metric, and SPPCSPC alone and Focal-EIoU alone are below the baseline. The sentence in Section 5.3.3 that "different improved steps can improve mAP_0.5, mAP_0.5:0.95, and recall" is therefore contradicted for mAP_0.5:0.95 by the SPPCSPC-only and Focal-EIoU-only rows. Without error bars or multiple seeds, these single-run values are as consistent with run-to-run variance as with a genuine interaction effect, so the claimed 4% improvement from the full combination is not established.
  2. [5.3.3 and 5.3.4, Tables 2 and 3] The paper reports exactly one run per configuration and provides no standard deviations, random seeds, or statistical tests. This is load-bearing because the headline differences are small (2.7% mAP_0.5, 4% mAP_0.5:0.95, 6% recall) while alternative configurations in Table 2 differ from the full model by more than 4 points on mAP_0.5:0.95. The authors should report mean and standard deviation over at least three seeds for the baseline, each ablated configuration, the full model, and the comparative models, or otherwise demonstrate that the observed ordering is stable rather than a single-run artifact.
  3. [3, Table 1] The dataset description is internally inconsistent. The text states that data augmentation results in "an expanded dataset comprising 11,323 images," but the "Data Augmentation" row of Table 1 sums to 13,773 labels (2024+1299+4017+2133+2489+1811). Since the table is titled "Total Amount of Each Type of Data," the reader cannot tell whether the entries are images, labeled instances, or something else. Please define the unit, reconcile the image count with the label count, and state explicitly how the 6:2:2 split was applied to images versus labels; Table 4's test-set label counts (1008 original and 2798 augmented) do not correspond to exactly 20% of the totals given in Table 1.
minor comments (6)
  1. [5.2] The metric name "GFloats" should be "GFLOPs"; the metric is defined correctly but the label is misspelled.
  2. [5.3.1] The caption describing the training curves contains the typo "SPPCSCP" where "SPPCSPC" is intended.
  3. [4.4, Equation (4)] Equation (4), the Focal-EIoU formula, is not legible in the submitted version; please re-typeset it and explicitly define the parameter gamma and the suppression mechanism.
  4. [3, Table 4] Please state explicitly that the "Labels" column in Table 4 refers to test-set instances, not total annotated instances, and clarify how images were assigned to the training, validation, and test splits given the stated 6:2:2 ratio.
  5. [5.3.1] The claim that the improved model "accelerates model convergence" is not supported by any quantitative measure such as epochs to a given mAP; the paper reports only the final training curves.
  6. [2.1 and References] The in-text citation for YOLOv7 points to reference [24] instead of [23], and subsequent citations to gradient-path analysis and ELAN appear to be shifted relative to the reference list; please correct the citation numbers.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the central claim is an empirical comparison against YOLOv8m and other detectors on a private dataset; the authors' earlier papers are cited only as related work and do not carry the derivation.

full rationale

The paper's central claim is that adding GAM, SPPCSPC, and Focal-EIoU to YOLOv8m improves mAP_0.5, mAP_0.5:0.95, and recall. This is supported by single-run evaluations in Tables 2 and 3, not by a derivation from an input definition. None of the components is defined in terms of the target metrics: GAM is an attention module from an external CVPR paper, SPPCSPC is a pooling module from YOLOv7-style architectures, and Focal-EIoU is a standard bounding-box loss. The paper does not fit a parameter to the test set and then rename the fit as a prediction; the improvements are stated as measured differences against YOLOv8m. The only self-citations are references [42]-[44] (same research group's previous YOLOv5s/YOLOv7-tiny power-line papers), used as examples of related transmission-line component/smoke detection work; they are not invoked to justify the proposed architecture or to forbid alternatives, so they are not load-bearing. The conclusion's limitation statement about the private dataset and planned release is a data-availability caveat, not circularity. I note for correctness, not circularity, that Table 2 is internally non-monotonic (GAM+Focal alone reaches mAP_0.5:0.95=84.5 while the full model reaches 80.4, and SPPCSPC or Focal-EIoU alone decrease the metric), so the attribution of the claimed 4% gain to the full combination is fragile; but this is an empirical-reliability concern, not a reduction of the claim to its inputs.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim is empirical; there are no free parameters fitted to data in the derivation sense. The training hyperparameters are standard and do not determine the qualitative result. No new entities are postulated.

assumptions (4)
  • domain assumption The private Yunnan Power Grid dataset is correctly labeled and representative.
    Dataset is not publicly available, so label quality and annotation consistency cannot be independently checked (Section 3).
  • domain assumption The 6:2:2 split of the augmented dataset keeps augmented variants of each source image in the same fold.
    Section 3 describes the ratio but does not state whether splitting is grouped by source image; without grouping, near-duplicate augmented images can leak between training and test.
  • domain assumption The cited YOLOv8m and module implementations are correct.
    No code is released; the comparison assumes the baseline and the three inserted modules are correctly integrated.
  • domain assumption Standard mAP, precision, and recall metrics capture the operational goal.
    The paper uses standard metrics (Section 5.2) but does not evaluate false-alarm cost or deployment constraints beyond speed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Foreign-Object Detection in High-Voltage Transmission Line Based on Improved YOLOv8m." pith.science (2026). https://pith.science/paper/HHF43BVO

@misc{pith2026250207175,
  author       = {Pith},
  title        = {Pith review of: Foreign-Object Detection in High-Voltage Transmission Line Based on Improved YOLOv8m},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HHF43BVO}},
  note         = {Machine review of arXiv:2502.07175}
}
read the original abstract

The safe operation of high-voltage transmission lines ensures the power grid's security. Various foreign objects attached to the transmission lines, such as balloons, kites and nesting birds, can significantly affect the safe and stable operation of high-voltage transmission lines. With the advancement of computer vision technology, periodic automatic inspection of foreign objects is efficient and necessary. Existing detection methods have low accuracy because foreign objects at-tached to the transmission lines are complex, including occlusions, diverse object types, significant scale variations, and complex backgrounds. In response to the practical needs of the Yunnan Branch of China Southern Power Grid Co., Ltd., this paper proposes an improved YOLOv8m-based model for detecting foreign objects on transmission lines. Experiments are conducted on a dataset collected from Yunnan Power Grid. The proposed model enhances the original YOLOv8m by in-corporating a Global Attention Module (GAM) into the backbone to focus on occluded foreign objects, replacing the SPPF module with the SPPCSPC module to augment the model's multiscale feature extraction capability, and introducing the Focal-EIoU loss function to address the issue of high- and low-quality sample imbalances. These improvements accelerate model convergence and enhance detection accuracy. The experimental results demonstrate that our proposed model achieves a 2.7% increase in mAP_0.5, a 4% increase in mAP_0.5:0.95, and a 6% increase in recall.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 39 canonical work pages

  1. [1]

    Real‐time Rendering Algorithm Optimization for Large‐scale Power Transmission Line Scenes Based on LOD

    Deng, Y.; Tang, Z.; Li, H.; Yang, Z. Real‐time Rendering Algorithm Optimization for Large‐scale Power Transmission Line Scenes Based on LOD. Comput. Mod. 2017, 257, 115–118

    work page 2017

  2. [2]

    A Comprehensive Review on Intelligent Processing of Visible Light Images for Power Over‐ head Line Inspection

    Liu, Z.; Liao, X.; Chen, J.; Jiao, H. A Comprehensive Review on Intelligent Processing of Visible Light Images for Power Over‐ head Line Inspection. Power Syst. Te c h n o l . 2020, 44, 1057–1069

    work page 2020

  3. [3]

    Improving YOLOv3’ s Foreign Object Detection Method for Transmission Lines

    Zhang, H.; Zhou, H.; Li, S.; Li, P. Improving YOLOv3’ s Foreign Object Detection Method for Transmission Lines. J. Laser 2022, 43, 82–87

    work page 2022

  4. [4]

    A Transmission Line Foreign Object Detection Method Based on Visual Saliency Analysis

    Wan, D.; Zhang, J.; Guo, X. A Transmission Line Foreign Object Detection Method Based on Visual Saliency Analysis. J. Te l e v. Te c h n o l . 2018, 42, 106–110

    work page 2018

  5. [5]

    Research on Foreign Object Detection on Transmission Lines Based on UAV Aerial Images

    Yu, G.; Zou, Z.; Fu, X.; Peng, J.; Shi, C. Research on Foreign Object Detection on Transmission Lines Based on UAV Aerial Images. Jiangxi Sci. 2022, 40, 223–228

    work page 2022

  6. [6]

    A Comprehensive Review on UAV‐based Smart Inspection Technology for Power Transmission Lines

    Liao, X.; Liu, Z.; Yan, Q. A Comprehensive Review on UAV‐based Smart Inspection Technology for Power Transmission Lines. J. Fuzhou Univ. 2020, 48, 198–209

    work page 2020

  7. [7]

    Unmanned Aerial Vehicle Intelligent Inspection System for Power Lines Based on Cloud‐Fog Edge Heterogeneous Collaboration

    Huang, Z.; Wang, Y.; Wang, H.; Gao, C.; Bai, C. Unmanned Aerial Vehicle Intelligent Inspection System for Power Lines Based on Cloud‐Fog Edge Heterogeneous Collaboration. China Electr. Power 2020, 53, 161–168

    work page 2020

  8. [8]

    Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmenta‐ tion

    Girshick, R.; Donahue, J.; Darrell, T.; Malik, J. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmenta‐ tion. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, 23–28 June 2014; pp. 580–587

    work page 2014

Show all 45 references
  1. [9]

    Fast R‐CNN

    Girshick, R. Fast R‐CNN. In Proceedings of the 2015 IEEE International Conference on Computer Vision and Pattern Recogni‐ tion, Santiago, Chile, 7–13 December 2015; pp. 1440–1448

    2015

  2. [10]

    Faster R‐CNN: Towards real‐time object detection with region proposal networks

    Ren, S.; He, k.; Girshick, R.; Sun, J. Faster R‐CNN: Towards real‐time object detection with region proposal networks. IEEE Tr a n s . Pattern Anal. Mach. Intell. 2017, 39, 1137–1149

    2017

  3. [11]

    You only look once: Unified,real‐time object detection

    Redmon, J.; Divvala, S.; Cirshick, R.; Farhadi, A. You only look once: Unified,real‐time object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 7–12 June 2015; pp. 779–788

    2015

  4. [12]

    SSD:single shot multibox detector.European Conference on Computer Vision and Pattern Recognition, Amsterdam, The Netherlands, 10–16 October 2016; pp

    Liu, W.; Anguelov , D.; Erhan, D.; Szegedy , C.; Reed, S.; Fu, C.; Berg, A. SSD:single shot multibox detector.European Conference on Computer Vision and Pattern Recognition, Amsterdam, The Netherlands, 10–16 October 2016; pp. 21–37

    2016

  5. [13]

    Girshick, R.; He, K.; Dollar, P

    Lin, T.; Goyal, P. ; Girshick, R.; He, K.; Dollar, P. Focal loss for dense object detection. In Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017; pp. 2980–2988

    2017

  6. [14]

    YOLO900:Better,faster,stronger

    Redmon, J.; Farhadi, A. YOLO900:Better,faster,stronger. In Proceeding of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017

    2017

  7. [15]

    YOLOv3:An incremental improve

    Redmon, J.; Farhadi, A. YOLOv3:An incremental improve. In Proceeding of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City , UT, USA, 18–22 June 2018. 24 of 24

    2018

  8. [16]

    Yolov4: Optimal speed and accuracy of object detection

    Bochkovskiy , A.; Wang, C.; Liao, H. Yolov4: Optimal speed and accuracy of object detection. arXiv 2020, arXiv:2004.10934

    2020 arXiv

  9. [17]

    YOLOv5 Object Detection

    ultralytics ultralytics. YOLOv5 Object Detection. [Online]. Available online: https://github.com/ultralytics/yolov5 (accessed on April 2020)

    2020

  10. [18]

    YOLOX: Exceeding YOLO Series in 2021

    Ge, Z.; Liu, S.; Wang, F.; Li, Z.; Sun, J. YOLOX: Exceeding YOLO Series in 2021. arXiv 2021, arXiv:2107.08430

    2021 arXiv

  11. [19]

    You Only Learn One Representation: Unified Network for Multiple Tasks

    Wang, C.; Yeh, I.; Liao, H. You Only Learn One Representation: Unified Network for Multiple Tasks. arXiv 2021, arXiv:2105.04206

    2021 arXiv

  12. [20]

    Li, C.; Li, L.; Jiang, H.; Weng, K.; Geng, Y.; Li, L.; Ke, Z.; Li, Q.; Cheng, M.; Nie, W. ; et al. YOLOv6: A Single‐Stage Object De‐ tection Framework for Industrial Applications. arXiv 2022, arXiv:2209.02976

    2022 arXiv

  13. [21]

    RepVGG: Making VGG‐style ConvNets Great Again

    Ding, X.; Zhang, X.; Ma, N.; Han, J.; Ding, G.; Sun, J. RepVGG: Making VGG‐style ConvNets Great Again. In Proceeding of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Online, 19–25 June 2021; pp. 13728–13737

    2021

  14. [22]

    SIoU Loss: More Powerful Learning for Bounding Box Regression

    Gevorgyan, Z. SIoU Loss: More Powerful Learning for Bounding Box Regression. arXiv 2022, arXiv:2205.12740

    2022 arXiv

  15. [23]

    YOLOv7: Trainable bag‐of‐freebies sets new state‐of‐the‐art for real‐time object detectors

    Wang, C.; Bochkovskiy , A.; Liao, H. YOLOv7: Trainable bag‐of‐freebies sets new state‐of‐the‐art for real‐time object detectors. In Proceeding of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Oxford, UK, 15–17 September 2023; pp. 7464–7475

    2023

  16. [24]

    Designing Network Design Strategies Through Gradient Path Analysis

    Wang, C.; Liao, H.; Yeh, I. Designing Network Design Strategies Through Gradient Path Analysis. arXiv 2022, arXiv:2211.04800

    2022 arXiv

  17. [25]

    YOLOv8 Object Detection

    Ultralytics. YOLOv8 Object Detection. [Online]. Available online: https://ultralytics.com/yolov8 (accessed on 17 July 2023)

    2023

  18. [26]

    Shao, Z.; Hoffmann, N

    Liu, Y. ; Shao, Z.; Hoffmann, N. Global Attention Mechanism: Retain Information to Enhance Channel‐Spatial Interactions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Online, 19–25 June 2021

    2021

  19. [27]

    Online Monitoring of Ice Thickness on Transmission Lines using Image Edge Detection

    Wang, X.; Hu, J.; Sun, C.; Du, L. Online Monitoring of Ice Thickness on Transmission Lines using Image Edge Detection. High Vo l t . Appar. 2009, 45, 69–73

    2009

  20. [28]

    Real‐Time Image‐Identification‐Based Anti‐Manmade Misoperation System for Substations

    Cai, W.; Le, J.; Jin, C.; Liu, K. Real‐Time Image‐Identification‐Based Anti‐Manmade Misoperation System for Substations. IEEE Tr a n s . Power Deliv. 2023, 27, 1748–1754

    2023

  21. [29]

    Recognition of Foreign Objects on Transmission Lines based on Aerial Images

    Jin, L.; Yao, C.; Yan, S.; Zhang, W. Recognition of Foreign Objects on Transmission Lines based on Aerial Images. J. To ng j i Univ. 2013, 41, 277–281

    2013

  22. [30]

    Zhang, J.; Han, J.; Liu, L.; Zhu, M

    Wang, W. ; Zhang, J.; Han, J.; Liu, L.; Zhu, M. Method for Detecting Line Breaks and Foreign Object Defects on Transmission Lines Based on Drone Images. J. Comput. Appl. 2015, 35, 2404–2408

    2015

  23. [31]

    Zhang, C.; Zheng, X

    Shi, P. ; Zhang, C.; Zheng, X. Foreign Object Detection on High‐voltage Poles based on Deep Learning. J. Chongqing Univ. Sci. Te c h n o l . 2020, 22, 83–87

    2020

  24. [32]

    Yue, F.; Yuan, H.; Rong, M

    Zhu, J.; Guo, Y. ; Yue, F.; Yuan, H.; Rong, M. A Deep Learning Method to Detect Foreign Objects for Inspecting Power Trans‐ mission Lines. IEEE Access 2020, 8, 94065–94075

    2020

  25. [33]

    A Method for Monitoring and Identifying Intrusions of Foreign Objects on Transmis‐ sion Lines based on Deep Learning Networks

    Yang, J.; Qin, Z.; Pang, X.; He, Z.; Cui, C. A Method for Monitoring and Identifying Intrusions of Foreign Objects on Transmis‐ sion Lines based on Deep Learning Networks. Power Syst. Prot. Control 2021, 49, 37–44

    2021

  26. [34]

    Qiu, Z.; Zhou, Y.; Zhu, X.; Wang, Q

    Yu, Y. ; Qiu, Z.; Zhou, Y.; Zhu, X.; Wang, Q. Transmission Line Foreign Object Detection Based on Convolutional Neural Network and ECOC‐SVM. Smart Power 2022, 50, 87–92+107

    2022

  27. [35]

    Qiu, Z.; Liao, H.; Wei, Z.; Zhu, X.; Zhou, Z

    Yu, Y. ; Qiu, Z.; Liao, H.; Wei, Z.; Zhu, X.; Zhou, Z. A Method Based on Multi‐Network Feature Fusion and Random Forest for Foreign Objects Detection on Transmission Lines. Appl. Sci. 2022, 12, 4982

    2022

  28. [36]

    An Improved YOLO Network for Detection of Small Foreign Objects on Trans‐ mission Lines

    Zou, H.; Jiao, L.; Zhang, Z.; Tang, B.; Liu, Z. An Improved YOLO Network for Detection of Small Foreign Objects on Trans‐ mission Lines. J. Nanjing Inst. Te c h n o l . 2022, 20, 7–14

    2022

  29. [37]

    Qiu, Z.; Zhu, X.; Liao, C.; Qu, W. ; Yu, Y. A Lightweight YOLOv4‐EDAM Model for Accurate and Real‐time Detection of Foreign Objects Suspended on Power Lines. IEEE Tr an s . Power Deliv. 2023, 38, 1329–1340

    2023

  30. [38]

    Wang, J.; Liu, M.; Kong, X.; Tang, J

    Wu, M.; Guo, L.; Chen, R.; Du, W. ; Wang, J.; Liu, M.; Kong, X.; Tang, J. Improved YOLOX Foreign Object Detection Algorithm for Transmission Lines. Wirel. Commun. Mob. Comput. 2022, 2022, 5835693

    2022

  31. [39]

    Zhang, W

    Yu, C.; Liu, Y. ; Zhang, W. ; Zhang, X.; Zhang, Y. ; Jiang, X. Foreign Objects Identification of Transmission Line Based on Improved YOLOv7. IEEE Access 2023, 11, 51997–52008

    2023

  32. [40]

    Detection of High Voltage Transmission Line Foreign Objects based on Improved YOLOv4 Algorithm

    Zhang, Q.; Zhu, T.; Xiao, S.; Yang, Z.; Zeng, H. Detection of High Voltage Transmission Line Foreign Objects based on Improved YOLOv4 Algorithm. Appl. Sci. 2023, 50, 59–65

    2023

  33. [41]

    Foreign object detection for transmission lines based on Swin Trans‐ former V2 and YOLOX

    Tang, C.; Dong, H.; Huang, Y.; Han, T.; Fang, M.; Fu, J. Foreign object detection for transmission lines based on Swin Trans‐ former V2 and YOLOX. Vis. Comput. 2023, 1432–2315. https://doi.org/10.1007/s00371‐023‐03004‐8

    2023 doi

  34. [42]

    Improved YOLOv5s model for key components detecion of power transmission lines

    Chen, C.; Yuan, G.; Zhou, H.; Ma, Y. Improved YOLOv5s model for key components detecion of power transmission lines. Math. Biosci. Eng. 2023, 20, 7738–7760

    2023

  35. [43]

    Transmission Lines Small‐Target Detection Algorithm Research Based on YOLOv5

    Cheng, Q.; Yuan, G.; Chen, D.; Xu, B.; Chen, E.; Zhou, H. Transmission Lines Small‐Target Detection Algorithm Research Based on YOLOv5. Appl. Sci. 2023, 13, 9386

    2023

  36. [44]

    Optimized YOLOv7‐tiny model for smoke detection in power transmission lines

    Chen, C.; Yuan, G.; Zhou, H.; Ma, Y. Optimized YOLOv7‐tiny model for smoke detection in power transmission lines. Math. Biosci. Eng. 2023, 20, 19300–19319

    2023

  37. [45]

    CBAM: Convolutional Block Attention Module

    Woo, S.; Park, J.; Lee, J.‐Y.; Kwoen, I.S. CBAM: Convolutional Block Attention Module. In Proceedings of the Springer Con‐ ference on European Conference on Computer Vision, Munich, Germany , 8–14 September 2018

    2018

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.