Browse by author
Lookup NU author(s): Sabrina SU, Professor Neil Boonham, Dr Huizhi LiangORCiD, Dr Ankush PrasharORCiD
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).
© 2025. Traditional machine learning and deep learning models have attracted wide attention for autonomous weed detection. Combinations of spectral, shape and textual features, together with shallow machine learning classifiers have demonstrated weed detection accuracy comparable to deep learning approaches. To leverage the strong interpretability and short training time of conventional machine learning techniques, this study proposed machine learning-based workflows, which relied solely on spectral features for weed discrimination, without extracting shape or textual features. Using multispectral datasets collected from the greenhouse and field, this study provides insights of the machine learning-based workflows performance under both controlled and uncontrolled conditions. The 10-band full spectrum outperformed discriminant bands selected by ReliefF, Recursive Feature Elimination and Principal Component Analysis across all classifiers. Among the models, K Nearest Neighbour achieved the highest performance, surpassing both Random Forest and Support Vector Machine, with F1 scores of 0.86 in the greenhouse and 0.96 in the field. The optimal timing for autonomous weeding was identified between the second leaf development and before side shoot formation (BBCH 12–19), corresponding to 52–65 days after sowing in the greenhouse and 64 days after sowing in the field. This recommendation was based on statistical analysis and the 10band-KNN model applied to multispectral data collected on different dates. These findings provide practical guidance for optimizing autonomous weeding schedules in real-world scenarios. Moreover, the results demonstrate the robustness of the proposed workflows across different weed species and environmental conditions, while also supporting the reliability of related studies conducted under controlled environments.
Author(s): Su R, Boonham N, Liang H, Prashar A
Publication type: Article
Publication status: Published
Journal: Smart Agricultural Technology
Year: 2025
Volume: 12
Print publication date: 01/12/2025
Online publication date: 11/07/2025
Acceptance date: 11/07/2025
Date deposited: 28/07/2025
ISSN (electronic): 2772-3755
Publisher: Elsevier BV
URL: https://doi.org/10.1016/j.atech.2025.101187
DOI: 10.1016/j.atech.2025.101187
Data Access Statement: Data will be made available on request.
Altmetrics provided by Altmetric
