Pith. sign in

REVIEW 4 major objections 5 minor 48 references

Towards Efficient and Intelligent Laser Weeding: Method and Dataset for Weed Stem Detection

T0 review · 4 major / 5 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 Adding a stem point head to a weed detector lifts simulated laser weeding accuracy by 6.7%.

desk verdict WSD dataset is the real contribution; the headline efficiency numbers rest on an unspecified simulation that needs fixing before publication. read the letter →

arxiv 2502.06255 v1 pith:NRK3NXD6 submitted 2025-02-10 cs.CV cs.AI

Dingning Liu , Jinzhe Li , Haoyang Su , Bei Cui , Zhihui Wang , Qingbo Yuan , Wanli Ouyang , Nanqing Dong This is my paper · ORCID
classification cs.CVcs.AI
keywords weedstemdetectionlaserweedingobjectYOLOv7localizationsemi-supervisedlearningprecisionagriculturedataset
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

This paper argues that for laser weeding, a weed is best targeted by its stem, and that existing detectors aimed at bounding-box centers miss this target. It introduces the Weed Stem Detection (WSD) dataset of 7,161 field-collected high-resolution images with human-annotated crop and weed boxes plus weed stem points, and adds a stem-coordinate regression head to YOLOv7. In the paper's simulated laser-weeding evaluation, the augmented model improves weeding accuracy from 75.37% to 80.42% and lowers energy cost from 1.55 to 1.05 unit shots, with zero crops misidentified as weeds. The result matters because laser weeding is proposed as a chemical-free weed control method, and precise stem targeting is what makes it energy-efficient and crop-safe.

What carries the argument

The central mechanism is an additional stem-coordinate regression head grafted onto a pretrained object detector, with a mean-squared-error loss $L_{reg}$ on weed stem coordinates combined as $L = \alpha L_{cls} + \beta L_{bbox} + \gamma L_{reg}$. This lets the same network output crop and weed boxes plus a precise stem point for each weed. The paper also extends the pipeline with a teacher-student semi-supervised loop that uses confidence scores and cosine similarity to a weed embedding bank to filter pseudo labels.

What would settle it

Define a distance threshold (for example, $\pm 2$ cm) and an optical and thermal kill model, then run the trained model on a real laser weeding platform under the same field conditions; if predicted stem points within threshold do not produce kill rates consistent with the simulated 80.42% accuracy and 1.05 shots per weed, the central claim would need revision.

Watch

Extended reading notes

Core claim

The paper claims that bounding-box geometric centers are unreliable proxies for weed stem locations, and that a dedicated stem regression branch fixes this without sacrificing detection. Integrating stem regression with object detection reduces stem localization distance error from 2.9770 to 2.4838, raises simulated weeding accuracy by 6.7%, and cuts the hypothetical minimum energy cost by 32.3%, all with a 0% false positive rate on crops misidentified as weeds. Adding semi-supervised training with pseudo-labeled unlabeled images further reduces the distance error to 2.1485.

Load-bearing premise

The evaluation assumes that hitting the annotated stem point with one laser shot in a simulation equals killing the weed in a real field; the paper does not specify the distance tolerance or the optical and thermal model that converts a predicted point into a successful kill.

Editorial extensions

If this is right

  • Using weed stem coordinates rather than box centers as the laser target translates to fewer laser shots and higher simulated kill rates.
  • The WSD dataset enables the first empirical comparison of weed recognition methods under a laser-weeding metric (Euclidean distance to the stem) rather than mAP alone.
  • Semi-supervised learning improves stem localization without new annotations, reducing the cost of scaling the approach to new fields.
  • Anchor-based YOLOv7 is reported to outperform anchor-free YOLOv8 and YOLOv10 for stem regression, suggesting anchor placement matters for point regression.
  • A zero false-positive rate on crops is reported across thresholds down to 0.056, indicating that the model can be run with a low detection threshold without misidentifying crops as weeds.

Reading between the lines

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

  • If the simulated one-shot model is replaced with a calibrated kill/no-kill criterion (laser power, dwell time, stem diameter), the reported accuracy and cost gains may change; the paper's numbers are best read as upper-bound-style estimates.
  • A natural extension is to test the regression head on weeds whose stems are occluded by leaves, since occlusion is common in real fields and the dataset currently does not separate visible from occluded stems.
  • The stem regression idea should transfer to other point-targeting agricultural tasks, such as fruit peduncle cutting or plant node pruning, where bounding box centers are also not the operative target.
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

4 major / 5 minor

Summary. The paper presents the Weed Stem Detection (WSD) dataset, a field-collected resource of 7,161 high-resolution images (1,556 annotated) containing crop/weed bounding boxes and weed stem point annotations, and proposes adding a stem-coordinate regression head to YOLOv7, plus a semi-supervised teacher-student extension that uses the 5,605 unannotated images. On the WSD test split, the method reduces the stem localization distance from 2.9770 to 2.4838, and to 2.1485 with semi-supervised learning. The authors further report a 'real-world simulated experiment' in which the method improves weeding accuracy from 75.37% to 80.42% and reduces energy cost from 1.55 to 1.05 unit shots. The paper claims to provide the first empirical study and the first human-annotated dataset for weed stem localization for laser weeding.

Significance. If the reported results hold, the WSD dataset fills a clear gap: existing weed datasets lack point-level stem annotations, which are precisely what laser weeding needs for energy-efficient aiming. The proposed modification is simple, end-to-end, and deployable on a lightweight detector, and the semi-supervised extension addresses the high annotation cost. The paper's strengths include a human-annotated dataset with consensus verification, a promised public code/data release, a held-out test split, and a zero-FP analysis for crops misidentified as weeds. However, the headline efficiency claims rest on an underspecified simulation metric, and the evaluation is single-run with hyperparameters tuned on the same dataset. These issues are fixable but prevent the current version from fully supporting the quantitative claims.

major comments (4)
  1. [Experiments: Results] Table 3 and the 'real-world simulated experiment' described under 'Effect of Stem Regression' do not specify the kill criterion. The reader is not told the Euclidean distance threshold (in pixels or physical units) below which a predicted stem counts as a successful laser kill, nor the laser spot size, aiming rule, or number of shots allowed per weed. The reported pairs (75.37%, 1.55) and (80.42%, 1.05) are not compatible with a simple one-shot-per-point model: if Accuracy were single-shot hit probability, the expected shots to first success would be 1/p (1.33 and 1.24), and if Accuracy were the fraction killed after unlimited shots it would be 100%. Some undisclosed multi-shot or energy-scaling rule is therefore used. Without this specification, the headline 6.7% accuracy gain and 32.3% energy saving cannot be reproduced or even interpreted; this is the most load-bearing weakness of the paper.
  2. [Experiments: Evaluation Metric] The evaluation metric Dist is the Euclidean distance between predicted and annotated stem coordinates, and the training loss Lreg in Eq. (3) is, by the paper's own definition, the same Euclidean distance on the same coordinates. The reported Dist comparisons therefore partly measure an optimization echo rather than independent predictive quality. This is compounded by the facts that all hyperparameters are 'determined on the basis of empirical results' on the same dataset and that the experiments appear to be single runs with no error bars, seeds, or cross-validation. I request repeated runs with variance reporting, or a clearly separated hyperparameter-selection protocol, before the Dist improvements are used as evidence of generalization.
  3. [Experimental Setup] The paper states that 80% of the data is used for training, 10% for validation, and 10% for testing, but the dataset contains 1,556 annotated and 5,605 unannotated images. It is unclear whether the unannotated images are included in the test split and whether the pseudo-labels for semi-supervised learning are generated from those 5,605 images. This matters because the SSL rows in Table 4 and Fig. 8 could involve test-set leakage if any unannotated images used for pseudo-labeling also appear in the test split. Please specify the exact assignment of annotated and unannotated images to the training, validation, and test splits, and confirm that the test set is never used for pseudo-label generation or threshold selection.
  4. [Ablation Study on Detection Backbone] The claim in Table 5 that YOLOv7's anchor-based design 'prevents drift by keeping loss calculation confined to the respective grid' is presented as an explanation, but YOLOv7, YOLOv8, and YOLOv10 differ in architecture, training schedule, and label assignment beyond the anchor-versus-anchor-free distinction. No controlled comparison isolates the anchor mechanism, so this statement overstates what the ablation can establish. A more cautious interpretation, or an experiment that isolates the anchor mechanism, is needed.
minor comments (5)
  1. [Weed Stem Detection Dataset, Table 1] The row for WSD reports average instances per image as 12.5 ± 7.5, but the text and Table 2 report 11,151 weed instances among 1,556 annotated images. Please clarify whether the average is computed over annotated images only, whether it includes crop instances, and align the notation with the text.
  2. [Method, Eq. (3)] The text says 'MSE(·) represents the Euclidean Distance calculation', but MSE is mean squared error, not Euclidean distance. If Eq. (3) truly uses MSE, the training loss is squared Euclidean distance, which is monotonic with Dist but not identical; please correct the wording.
  3. [Figure 8] The caption and text claim that semi-supervised learning is model agnostic, but the figure appears to show only two backbones and no tabulated values, confidence intervals, or repeated runs; please add quantitative results to support this claim.
  4. [Table 3] The caption defines unit cost as the 'hypothetical minimum energy cost with ground truth-level prediction, i.e one weed only requires one shot', yet the reported costs are 1.55 and 1.05. Please reconcile this definition with the simulation rule, or reword the caption to describe the actual energy model.
  5. [Ablation Study on Detection Backbone] There is a typo in the section heading: 'Peformance' should be 'Performance'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: training and evaluation share the Euclidean distance metric, but the reported gains are computed on held-out human annotations and no load-bearing step reduces to its own inputs.

full rationale

The paper's derivation chain is self-contained. The regression loss Lreg (Eq. 3) is MSE on weed stem coordinates, and the evaluation metric Dist is Euclidean distance between predicted and ground-truth stem points. Sharing a metric between training and evaluation is not circular, because the reported numbers are computed on a held-out test split with independent human annotations; the method is not fitting parameters to the test labels. The simulated weeding accuracy and energy cost (Table 3) are derived from the same stem-coordinate predictions, and the paper does not specify the distance threshold or laser/energy model underlying the simulation, which is a reproducibility and validity concern for the headline efficiency claim rather than a circularity. There are no load-bearing self-citations: YOLOv7 and SSL baselines are cited from external work, no uniqueness theorem is invoked, and the WSD dataset claim is a novelty claim about annotation content, not a result derived from itself. The possible optimization echo, where training on MSE improves Dist, is an expected and legitimate experimental outcome, not a reduction of the claim to its inputs.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

No new physical entities are postulated. The central claims rest on five tuned hyperparameters and four domain assumptions, chiefly that a single stem point and the simulated one-shot evaluation capture real laser weeding behavior.

free parameters (5)
  • Loss weights alpha, beta, gamma = 0.2, 0.3, 0.5
    Chosen empirically per the paper to balance classification, box, and stem regression losses.
  • SSL confidence threshold tau = 0.5
    Pseudo-labels are accepted only above this confidence; chosen empirically.
  • SSL cosine similarity threshold xi = 0.4
    Filters pseudo-label weed embeddings against the labeled weed bank; chosen empirically.
  • EMA smoothing factor = 0.9
    Teacher update smoothing in semi-supervised learning; chosen empirically.
  • Inference confidence threshold = 0.15
    Chosen to match zero-shot object detection conventions; Table 6 shows the minimum zero-FP threshold on the data is 0.056.
assumptions (4)
  • domain assumption A single 2D point in a top-down RGB image is a sufficient target for laser stem cutting.
    The entire dataset and evaluation encode weed stems as one coordinate; laser aiming accuracy, thermal kill radius, and stem occlusion are not modeled. This enters in Data Annotation and the simulated experiment.
  • domain assumption The simulated one-shot-per-weed evaluation predicts real weeding outcomes.
    Accuracy and energy cost in Table 3 are derived from predicted versus ground-truth Euclidean distance without a specified kill threshold or laser model. Invoked in Experiments: Results.
  • domain assumption Agronomist point annotations are treated as ground truth.
    No inter-annotator agreement statistics are reported; consensus was achieved by discussion and re-annotation. Invoked in Data Annotation.
  • ad hoc to paper YOLOv7 anchors confine loss calculation to the grid, explaining why anchor-based models outperform anchor-free models.
    This explanation is offered as a hypothesis with citations to blog posts, not a derivation or controlled experiment. Invoked in Experiments: Ablation Study on Detection Backbone.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Efficient and Intelligent Laser Weeding: Method and Dataset for Weed Stem Detection." pith.science (2026). https://pith.science/paper/NRK3NXD6

@misc{pith2026250206255,
  author       = {Pith},
  title        = {Pith review of: Towards Efficient and Intelligent Laser Weeding: Method and Dataset for Weed Stem Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NRK3NXD6}},
  note         = {Machine review of arXiv:2502.06255}
}
read the original abstract

Weed control is a critical challenge in modern agriculture, as weeds compete with crops for essential nutrient resources, significantly reducing crop yield and quality. Traditional weed control methods, including chemical and mechanical approaches, have real-life limitations such as associated environmental impact and efficiency. An emerging yet effective approach is laser weeding, which uses a laser beam as the stem cutter. Although there have been studies that use deep learning in weed recognition, its application in intelligent laser weeding still requires a comprehensive understanding. Thus, this study represents the first empirical investigation of weed recognition for laser weeding. To increase the efficiency of laser beam cut and avoid damaging the crops of interest, the laser beam shall be directly aimed at the weed root. Yet, weed stem detection remains an under-explored problem. We integrate the detection of crop and weed with the localization of weed stem into one end-to-end system. To train and validate the proposed system in a real-life scenario, we curate and construct a high-quality weed stem detection dataset with human annotations. The dataset consists of 7,161 high-resolution pictures collected in the field with annotations of 11,151 instances of weed. Experimental results show that the proposed system improves weeding accuracy by 6.7% and reduces energy cost by 32.3% compared to existing weed recognition systems.

Figures

Figures reproduced from arXiv: 2502.06255 by the authors.

Figure 1
Figure 1. (a) Weed detection with YOLOv7: Yellow dashed [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Distribution of instance annotations per image: [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 2
Figure 2. Image samples show raw images (left) and 16x [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The pipeline of intelligent laser weeding. The autonomous vehicle captures the image (or video frame). The proposed [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Overview of semi-supervised learning process. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison of weed detection results between the vanilla detection method and our method. The results of [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 8
Figure 8. Figure 8: Effect of semi-supervised learning. Semi [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 7
Figure 7. Figure 7: Qualitative comparison of stem localization results [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 43 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    J.; Pettersen, T.; and Borchert, A

    Aamlid, T.; Hessels e, K. J.; Pettersen, T.; and Borchert, A. F. 2021. ROBO-GOLF: Robotic mowers for better turf quality on golf course fairways and semi-roughs, Results from 2020. NIBIO Rapport

    work page 2021

  4. [4]

    Berthelot, D.; Carlini, N.; Goodfellow, I.; Papernot, N.; Oliver, A.; and Raffel, C. A. 2019. Mixmatch: A holistic approach to semi-supervised learning. Advances in neural information processing systems, 32

    2019

  5. [5]

    Carbon Robotics . 2022. Laserweeder Implement

    work page 2022

  6. [6]

    Chen, D.; Lu, Y.; Li, Z.; and Young, S. 2022. Performance evaluation of deep transfer learning on multi-class identification of common weed species in cotton production systems. Computers and Electronics in Agriculture, 198: 107091

    work page 2022

  7. [7]

    Dang, F.; Chen, D.; Lu, Y.; and Li, Z. 2023. YOLOWeeds: A novel benchmark of YOLO object detectors for multi-class weed detection in cotton production systems. Computers and Electronics in Agriculture, 205: 107655

    work page 2023

  8. [8]

    D.; Imperoli, M.; Pretto, A.; and Nardi, D

    Fawakherji, M.; Potena, C.; Bloisi, D. D.; Imperoli, M.; Pretto, A.; and Nardi, D. 2019. UAV image based crop and weed distribution estimation on embedded GPU boards. In Computer Analysis of Images and Patterns: CAIP 2019 International Workshops, ViMaBi and DL-UAV, Salerno, Italy, September 6, 2019, Proceedings 18, 100--108. Springer

    work page 2019

Show all 48 references
  1. [9]

    M.; Sohel, F.; Diepeveen, D.; Laga, H.; and Jones, M

    Hasan, A. M.; Sohel, F.; Diepeveen, D.; Laga, H.; and Jones, M. G. 2021. A survey of deep learning techniques for weed detection from images. Computers and Electronics in Agriculture, 184: 106067

    2021

  2. [10]

    Hu, K.; Wang, Z.; Coleman, G.; Bender, A.; Yao, T.; Zeng, S.; Song, D.; Schumann, A.; and Walsh, M. 2024. Deep learning techniques for in-crop weed recognition in large-scale grain production systems: a review. Precision Agriculture, 25(1): 1--29

    2024

  3. [11]

    Y.; Tian, L

    Jeon, H. Y.; Tian, L. F.; and Zhu, H. 2011. Robust crop and weed segmentation under uncontrolled outdoor illumination. Sensors, 11(6): 6270--6283

    2011

  4. [12]

    Jiang, H.; Zhang, C.; Qiao, Y.; Zhang, Z.; Zhang, W.; and Song, C. 2020. CNN feature based graph convolutional network for weed and crop recognition in smart farming. Computers and Electronics in Agriculture, 174: 105450

    2020

  5. [13]

    Joiya, F. 2022. YOLO vs. Faster R-CNN for Object Detection. International Research Journal of Modernization in Engineering Technology and Science (IRJMETS). Accessed: 2024-08-14

    2022

  6. [14]

    Keylabs. 2023. YOLOv8 vs SSD: Choosing the Right Object Detection Model. Keylabs. Accessed: 2024-08-14

    2023

  7. [15]

    P.; Mohamed, S.; Jimenez Rezende, D.; and Welling, M

    Kingma, D. P.; Mohamed, S.; Jimenez Rezende, D.; and Welling, M. 2014. Semi-supervised learning with deep generative models. Advances in neural information processing systems, 27

    2014

  8. [16]

    E.; Fu, C.; and Berg, A

    Liu, W.; Anguelov, D.; Erhan, D.; Szegedy, C.; Reed, S. E.; Fu, C.; and Berg, A. C. 2016. SSD: Single Shot MultiBox Detector. In European Conference on Computer Vision, 21--37

    2016

  9. [17]

    Long, J.; Shelhamer, E.; and Darrell, T. 2015. Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition, 3431--3440

    2015

  10. [18]

    Lu, Y. 2023. CottonWeedDet12: A 12-class weed dataset of cotton production systems for benchmarking AI models for weed detection. Zenodo

    2023

  11. [19]

    Nanos, G. 2023. Object Detection: SSD Vs. YOLO. Baeldung on Computer Science. Accessed: 2024-08-14

    2023

  12. [20]

    Nasiri, A.; Omid, M.; Taheri-Garavand, A.; and Jafari, A. 2022. Deep learning-based precision agriculture through weed recognition in sugar beet fields. Sustainable computing: Informatics and systems, 35: 100759

    2022

  13. [21]

    A.; Philippa, B.; Ridd, P.; Wood, J

    Olsen, A.; Konovalov, D. A.; Philippa, B.; Ridd, P.; Wood, J. C.; Johns, J.; Banks, W.; Girgenti, B.; Kenny, O.; Whinney, J.; et al. 2019. DeepWeeds: A multiclass weed species image dataset for deep learning. Scientific Reports, 9(1): 2058

    2019

  14. [22]

    V.; and Lloret, J

    Parra, L.; Marin, J.; Yousfi, S.; Rinc \'o n, G.; Mauri, P. V.; and Lloret, J. 2020. Edge detection for weed recognition in lawns. Computers and Electronics in Agriculture, 176: 105684

    2020

  15. [23]

    Pirchio, M.; Fontanelli, M.; Frasconi, C.; Martelloni, L.; Raffaelli, M.; Peruzzi, A.; Caturegli, L.; Gaetani, M.; Magni, S.; Volterrani, M.; et al. 2018. Autonomous rotary mower versus ordinary reel mower—effects of cutting height and nitrogen rate on Manila grass turf qualit...

    2018

  16. [24]

    Rahman, A.; Lu, Y.; and Wang, H. 2023. Performance evaluation of deep learning object detectors for weed detection for cotton. Smart Agricultural Technology, 3: 100126

    2023

  17. [25]

    Reis, D.; Kupec, J.; Hong, J.; and Daoudi, A. 2023. Real-time flying object detection with YOLOv8. arXiv preprint arXiv:2305.09972

    2023 arXiv

  18. [26]

    Ren, S.; He, K.; Girshick, R.; and Sun, J. 2016. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6): 1137--1149

    2016

  19. [27]

    Roboflow. 2023. YOLOv8 vs. Faster R-CNN: Compared and Contrasted. Accessed: 2024-08-14

    2023

  20. [28]

    M.; Schumann, A

    Sharpe, S. M.; Schumann, A. W.; Yu, J.; and Boyd, N. S. 2020. Vegetation detection and discrimination within vegetable plasticulture row-middles using a convolutional neural network. Precision Agriculture, 21: 264--277

    2020

  21. [29]

    Sportelli, M.; Pirchio, M.; Fontanelli, M.; Volterrani, M.; Frasconi, C.; Martelloni, L.; Caturegli, L.; Gaetani, M.; Grossi, N.; Magni, S.; et al. 2020. Autonomous mowers working in narrow spaces: A possible future application in agriculture? Agronomy, 10(4): 553

    2020

  22. [30]

    Steininger, D.; Trondl, A.; Croonen, G.; Simon, J.; and Widhalm, V. 2023. The cropandweed dataset: A multi-modal learning approach for efficient crop and weed manipulation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 3729--3738

    2023

  23. [31]

    Sudars, K.; Jasko, J.; Namatevs, I.; Ozola, L.; and Badaukis, N. 2020. Dataset of annotated food crops and weed images for robotic computer vision control. Data in brief, 31: 105833

    2020

  24. [32]

    Tanveer, A.; Chaudhry, N.; Ayub, M.; and Ahmad, R. 2003. Effect of cultural and chemical weed control methods on weed population and yield of cotton. Pak. J. Bot, 35(2): 161--166

    2003

  25. [33]

    Tao, T.; and Wei, X. 2022. A hybrid CNN--SVM classifier for weed recognition in winter rape field. Plant Methods, 18(1): 29

    2022

  26. [34]

    United Nations . 2023. Leave no one behind. https://unsdg.un.org/2030-agenda/universal-values/leave-no-one-behind

    2023

  27. [35]

    N.; Li, J.; Scott, S.; Psota, E.; J

    Veeranampalayam Sivakumar, A. N.; Li, J.; Scott, S.; Psota, E.; J. Jhala, A.; Luck, J. D.; and Shi, Y. 2020. Comparison of object detection and patch-based classification deep learning models on mid-to late-season weed detection in UAV imagery. Remote Sensing, 12(13): 2136

    2020

  28. [36]

    Wang, A.; Chen, H.; Liu, L.; Chen, K.; Lin, Z.; Han, J.; and Ding, G. 2024. Yolov10: Real-time end-to-end object detection. arXiv preprint arXiv:2405.14458

    2024 arXiv

  29. [37]

    Wang, C.-Y.; Bochkovskiy, A.; and Liao, H.-Y. M. 2023. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 7464--7475

    2023

  30. [38]

    WeLASER . 2023. WeLASER Implement

    2023

  31. [39]

    Wu, Z.; Chen, Y.; Zhao, B.; Kang, X.; and Ding, Y. 2021. Review of weed detection methods based on computer vision. Sensors, 21(11): 3647

    2021

  32. [40]

    Xu, M.; Zhang, Z.; Hu, H.; Wang, J.; Wang, L.; Wei, F.; Bai, X.; and Liu, Z. 2021. End-to-end semi-supervised object detection with soft teacher. In Proceedings of the IEEE/CVF international conference on computer vision, 3060--3069

    2021

  33. [41]

    You, J.; Liu, W.; and Lee, J. 2020. A DNN-based semantic segmentation for detecting weed and crop. Computers and Electronics in Agriculture, 178: 105750

    2020

  34. [42]

    W.; Cao, Z.; Sharpe, S

    Yu, J.; Schumann, A. W.; Cao, Z.; Sharpe, S. M.; and Boyd, N. S. 2019. Weed detection in perennial ryegrass with deep learning convolutional neural network. Frontiers in Plant Science, 10: 1422

    2019

  35. [43]

    Zhai, X.; Oliver, A.; Kolesnikov, A.; and Beyer, L. 2019. S4l: Self-supervised semi-supervised learning. In Proceedings of the IEEE/CVF international conference on computer vision, 1476--1485

    2019

  36. [44]

    Zhang, H.; Cao, D.; Zhou, W.; and Currie, K. 2024. Laser and optical radiation weed control: a critical review. Precision Agriculture, 1--25

    2024

  37. [45]

    Zhang, H.; Zhong, J.-X.; and Zhou, W. 2023. Precision optical weed removal evaluation with laser. In CLEO: Applications and Technology, JW2A--145. Optica Publishing Group

    2023

  38. [46]

    Zhang, L.; Zhang, Z.; Wu, C.; and Sun, L. 2022. Segmentation algorithm for overlap recognition of seedling lettuce and weeds based on SVM and image blocking. Computers and Electronics in Agriculture, 201: 107284

    2022

  39. [47]

    Zhang, Z. 1996 a . Developing chemical weed control and attaching importance to integrated weed management. In Proceedings of the National Symposium of IPM in China, 1996. China Agricultural Science and Technology Press

    1996

  40. [48]

    Zhang, Z. 1996 b . Weeds and their control in cotton fields, volume 2, 1345--1349. Beijing: China Agriculture Press

    1996

Pith tools

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