Abstract: With the advancement of precision agriculture, efficient and accurate weed detection has emerged as a pivotal task in modern crop management. Current weed detection methods face dual challenges: inadequate extraction of detailed features and edge information, coupled with the necessity for real-time performance. To address these issues, this paper pro-poses a lightweight multi-scale weed detection model based on YOLOv11n (You-only-look-once-11). Our approach incorporates three innovative components: (1) A fast-gated lightweight unit combined with C3K2 to enhance local and global interaction capabilities of weed features. (2) An adaptive hierarchical feature fusion network based on HSFPN, which improves the extraction of weed edge information. (3) A lightweight group convolution detection head module that captures multi-scale feature details while maintaining a lightweight structure. Experimental validation on two public datasets, CottonWeedDet3 and CottonWeed2, demonstrates that our model achieves an mAP50 improvement of 2.5% on CottonWeedDet3 and 1.9% on CottonWeed2 compared to YOLOv11n, with a 37% reduction in parameters and a 26%decrease in computational effort.

YaJun Zhang, Yu Xu, Jie Hou and YanHai Song, “LMS-YOLO11n: A Lightweight Multi-Scale Weed Detection Model” International Journal of Advanced Computer Science and Applications(IJACSA), 16(1), 2025. http://dx.doi.org/10.14569/IJACSA.2025.01601123

@article{Zhang2025,
title = {LMS-YOLO11n: A Lightweight Multi-Scale Weed Detection Model},
journal = {International Journal of Advanced Computer Science and Applications},
doi = {10.14569/IJACSA.2025.01601123},
url = {http://dx.doi.org/10.14569/IJACSA.2025.01601123},
year = {2025},
publisher = {The Science and Information Organization},
volume = {16},
number = {1},
author = {YaJun Zhang and Yu Xu and Jie Hou and YanHai Song}
}